the trading volume impact of local bias: evidence from a ... & weber...the trading volume impact...
TRANSCRIPT
Electronic copy available at: http://ssrn.com/abstract=1527579Electronic copy available at: http://ssrn.com/abstract=1527579
The Trading Volume Impact of Local Bias: Evidence from a
Natural Experiment
Heiko Jacobs and Martin Weber∗
June 1, 2011
Review of Finance, forthcoming
Appendix attached
Abstract
Exploiting regional holidays in Germany as a source of exogenous cross-sectional variation in in-vestor attention, we provide evidence that the well-known local bias at the individual level materiallyaffects stock turnover at the firm level. Stocks of firms located in holiday regions are temporarily strik-ingly less traded than otherwise very similar stocks in non-holiday regions. This negative turnovershock survives comprehensive tests for differences in information release. It appears particularly pro-nounced in stocks less visible to non-local investors, and for smaller stocks disproportionately drivenby retail investors. Our findings contribute to research on local bias, trading activity, and investordistraction.
Keywords: Local bias, trading activity, investor distraction, holiday effect, natural experiment
JEL Classification Codes: G12, G14
∗Heiko Jacobs is from the Lehrstuhl fur Bankbetriebslehre, Universitat Mannheim, L 5, 2, 68131 Mannheim. E-Mail:
[email protected]. Martin Weber is from the Lehrstuhl fur Bankbetriebslehre, Universitat Mannheim, L
5, 2, 68131 Mannheim, and CEPR, London. E-Mail: [email protected]. Mark Seasholes and an anonymous
referee provided many constructive suggestions that considerably improved the paper. We thank seminar participants at
the 17th Annual Meeting of the German Finance Association (DGF), at the 9th Cologne Colloquium on Financial Markets,
at the 14th Conference of the Swiss Society for Financial Market Research (SGF), and at the University of Mannheim for
valuable comments. We would like to thank Alexander Hillert for excellent research assistance. Financial support from the
Deutsche Forschungsgemeinschaft (SFB 504) and the Hamburg Stock Exchange is gratefully acknowledged.
1
Electronic copy available at: http://ssrn.com/abstract=1527579Electronic copy available at: http://ssrn.com/abstract=1527579
1. Introduction
By now there is ample evidence that both private and professional investors have a strong
preference for trading stocks of locally-headquartered firms. But is this so-called local bias
strong and pervasive enough to matter for the cross-section of stock turnover at the firm
level? To answer this question, we run a natural experiment in the German stock market.
Germany has several holidays which are observed only in some of its 16 states. While
these holidays have a religious origin, they materially influence public life as a whole.
Authorized by law, they are characterized by a limit or ban on work and official business
(but not exchanges). Previous research (e.g. DellaVigna and Pollet (2009), Hong and Yu
(2009), and Frieder and Subrahmanyam (2004)) and casual evidence suggest that both
private and professional investors in holiday regions tend to be temporarily distracted and
thus to often refrain from actively participating in the stock market on such days.
This exogenous variation in investor attention along a geographical line would not have
implications for the cross-section of abnormal firm-level trading activity if investors traded
the market portfolio. Only the aggregate level of trading volume might then be affected
(e.g. Lo and Wang (2000)). However, the introduction of local bias gives rise to a cross-
sectional hypothesis untested so far: Stocks of firms located in holiday regions (in the
following referred to as holiday firms) should, all else equal, exhibit a more pronounced
negative shock in trading activity than stocks of firms located in unaffected regions (in the
following referred to as non-holiday firms). An advantage of the German setting is that
both samples are similar and thus satisfy the requirements of a natural experiment: They
are broadly homogenous with respect to e.g. the number of firms, industry composition,
typical firm size, average stock risk-return profiles or (unconditional) turnover properties.
Similar findings apply to important characteristics of individual investors.
Consistent with our line of reasoning, we indeed find that holiday firms are (only) tem-
porarily strikingly less traded, both in statistical and economic terms. The negative shock
in turnover relative to non-holiday firms ranges roughly from 10% to 20%. It is not affected
by the inclusion of various control variables or several changes in methodology.
To the extent that news arrival triggers abnormal trading, one might be concerned that
2
our findings could be driven by a temporary change in the cross-section of information
release. Note, however, that the vast amount of firm-relevant news on a market, industry,
style or other aggregated levels should not be affected by regional holidays. It is arguably
only the structure of idiosyncratic firm-specific news, generated in or near a firm’s head-
quarter, which might potentially be affected. Digging deeper, we explore this news-based
explanation of our findings from five perspectives. From a firm perspective, we analyze
shocks in the release of corporate news. From a market perspective, we study shocks in
the idiosyncratic component of stock returns. From an investor’s viewpoint, we explore
shocks in the search frequency for firm names in Google. From an analyst perspective,
we study shocks in the cross-section of stock recommendations. From a media point of
view, we analyze shocks in press coverage. Overall, these tests (only) sporadically point
to significant differences in information release. Thus, we cannot rule out the possibility
of lower information intensity for holiday firms contributing to our results. However, we
believe it is justified to argue that information effects are unlikely to fully explain the
magnitude and robustness of the findings we document.
In line with a local bias explanation and the investor recognition hypothesis of Merton
(1987), the regional holiday effect is particularly pronounced for firms less visible to non-
local investors. Market capitalization, idiosyncratic risk and residual media coverage are
used as proxies for visibility. Finally, we study daily trading patterns of about 3,000
private investors from a German online broker. Consistent with implications of previous
research, individual investors seem to disproportionately cause the negative turnover shock
in smaller firms, in which their localized trading is concentrated.
Our study contributes to the literature in several ways. First, while prior research shows
that investors are biased towards the stocks of nearby firms, we identify scenarios in which
these individual preferences are strong and pervasive enough to materially affect the cross-
section of stock turnover. To our knowledge, our novel approach thereby provides the first
non-US evidence of local bias affecting market outcomes.
Second, our findings help to better understand determinants of stock-level trading volume,
which plays an essential role in much research on liquidity, return predictability, behavioral
finance or information asymmetries. For example, Hong and Stein (2007) note that “many
of most interesting patterns in prices and returns are tightly linked to movements in
3
volume” (p. 111). At the same time, empirical evidence on the drivers of its substantial
variation both in the cross-section and time-series is scarce (see e.g. the discussions in Gao
and Lin (2010), Statman et al. (2006) or Chordia et al. (2007)). We add to this literature by
uncovering cross-sectional regularities related to firm location, firm visibility, and investor
clienteles.
Third, a growing body of research builds on the idea of limited attention, whereby in-
vestors process only a subset of publicly available information due to attention capacity
constraints. A challenge for empirical work is the identification of a suitable proxy for
investor distraction. For example, Hou et al. (2009) rely on down market periods, while
Hirshleifer et al. (2009) employ the number of competing earnings announcements. In a
scenario related to ours, DellaVigna and Pollet (2009) analyze the market response to
earnings announcements on Fridays, when, as they argue, investor inattention is more
likely. Our findings highlight the role of regional holidays as a promising proxy for limited
attention. We identify scenarios which seem to cause distraction of an important sub-
set of investors, leading to market frictions in trading activity along a geographical line.
Moreover, we explore which firms and investor groups tend to be most affected.
The remainder of this paper is organized as follows. Section two discusses related research
and develops our hypotheses. Section three describes sample characteristics. Section four
contains the event study and explores alternative interpretations of our findings. Section
five analyzes determinants of the regional holiday effect. Section six concludes.
4
2. Related Literature and Hypotheses
By now, there is extensive and robust evidence for local bias on an individual level.1
However, research exploring its implications for return and volume patterns is still at the
beginning and moreover limited to the US market. Pirinsky and Wang (2006) document
an excessive comovement of local stock returns, which they attribute to correlated trad-
ing of local residents. Building on investors’ consumption smoothing motives, Korniotis
and Kumar (2010) argue that stock returns contain a predictable local component. The
findings of Hong et al. (2008) suggest that, in the presence of only few local firms compet-
ing for investors’ money, share prices of spatially close firms are driven up by the excess
demand of proximate residents. In a current study based on intra-day data, Shive (2011)
exploits large power outages to study the effect of local investor clienteles on pricing effi-
ciency. Her study provides evidence that informed local investors play an important role
in information processing and price discovery.
To the best of our knowledge, only two papers focus on the impact of local bias on firm-
level turnover. Loughran and Schultz (2004) show, among other pieces of evidence, that
the time zone in which a firm is headquartered triggers intraday trading patterns in its
stock. Loughran and Schultz (2005) demonstrate that rural stocks are less liquid than
urban stocks, which they attribute to the latter being local and thus visible to more
potential investors. They conclude that “much remains to be done on geography and
asset pricing” (p. 363). We aim at taking a step in this direction by exploiting holidays
which are observed only in some areas of Germany. In our baseline analysis, we focus
on All Saints’ Day as well as on Epiphany. All Saints’ Day, celebrated on November
1, is legally recognized only in the states of Baden-Wuerttemberg, Bavaria, Northrhine-
1Heterogeneous findings suggest that both informational and behavioral factors are likely to drive local bias. Studies
attributing this behavior to a preference for investing into the familiar, to the pronounced visibility of local stocks or to
incorrectly perceived information advantages include e.g. Bailey et al. (2008), Grinblatt and Keloharju (2001), Huberman
(2001), Seasholes and Zhu (2010), and Zhu (2003). Papers arguing in favor of superior locally generated information include
e.g. Baik et al. (2010), Bodnaruk (2009), Coval and Moskowitz (1999), Coval and Moskowitz (2001), Feng and Seasholes
(2004), Ivkovic and Weisbenner (2005), and Massa and Simonov (2006). Moreover, recent studies of Brown et al. (2008),
Hong et al. (2004), Hong et al. (2005), Ivkovic and Weisbenner (2007), and Shive (2010) show that local social interaction
and neighborhood word-of-mouth effects strongly affect investment decisions. Local bias has been shown to be robust
across countries, investor subgroups and sample periods. For the German market, combined findings from e.g. Dorn and
Huberman (2005), Dorn et al. (2008), Hau (2001), and this study suggest that, in the overall picture, German investors
pose no exception.
5
Westphalia, Rhineland Palatinate, and Saarland. Epiphany, celebrated on January 6, is a
legally recognized holiday only in the states of Baden-Wuerttemberg, Bavaria, and Saxony-
Anhalt. There are more regional celebrations in Germany (see the appendix, which is
available as supplemental material on the journal’s homepage www.revfin.org and from
the publisher web site rof.oxfordjournals.org). We partly rely on these holidays in later
tests. However, focusing on Epiphany and All Saints’ Day yields the most attractive event
study properties: It is a yearly event which splits the market in two large disjunct groups
with similar characteristics (see section 3 for details).
How holidays in general affect (in particular private) investors’ trading behavior is an
empirical question. On the one hand, one might expect increased trading activity, as in-
vestors may have more time to engage in the stock market. On the other hand, one might
expect decreased trading activity, as investors could indulge in vacation activities and
thus refrain from participation in the market. Indeed, previous work supports this second
line of reasoning. Frieder and Subrahmanyam (2004) show that turnover drops during
nationwide holidays. Hong and Yu (2009) provide evidence of aggregate trading activity
in international stock markets (including Germany) being lower during summer holiday
periods, which they dub a “gone fishin’ effect”. This seasonality in turnover seems to
be caused by both private and professional investors. DellaVigna and Pollet (2009) re-
port that trading activity immediately after earnings announcements made on Fridays is
comparatively low, as investors tend to be absent-minded due to the upcoming weekend.
With regard to the German setting, the idea of investors being temporarily distracted is
backed up by anecdotal evidence from leading papers and news services.2 Combined with
local bias, this type of limited investor attention makes novel predictions. Specifically,
if investors tend to heavily overweight local stocks in their investment decisions, then a
large, geographically concentrated subset of holiday-distracted investors might temporar-
2For instance, Die Welt (May 27, 2005), Financial T imes Deutschland (June 12, 2009), Tagesspiegel (June 12, 2009),
Stuttgarter Zeitung (May 8, 2007; May 24, 2008), DPA (May 25, 2001), and Dow Jones (June 1, 2007) all report that
many investors, both private and professional, stay out of the market on regional holidays and corresponding bridge days.
Other articles indirectly point to (primarily retail) investor distraction. For example, Frankfurter Rundschau (October 30,
2004) and Die Welt (November 2, 2004) report that non-holiday states profit from increased holiday tourism. AHGZ (May
12, 2007), a magazine for the hotel and catering sector, states that retail sales volume is higher around regional holidays.
Spiegel Online (June 14, 2006) and ddp (June 8, 2009) point to the danger of traffic jams due to the large number of people
on a short holiday. Sueddeutsche Zeitung (October 31, 2000) writes about massive obstructions of traffic near graveyards
on All Saints’ Day, on which it is custom to honor the deceased.
6
ily change the cross-section of stock turnover:
Hypothesis 1: Due to local bias, trading activity during regional holidays will be significantly
lower for firms in holiday regions.
This hypothesis is consistent with the trading volume implications of the habitat-based
model of comovement in Barberis et al. (2005). Similarly, in the model of Merton (1987),
investors are aware only of a subset of the stock universe. Consequently, the demand for
each stock depends on its shadow cost of information. In equilibrium, firms recognized by
less investors, will, all else equal, have fewer shareholders taking relatively large positions.
It seems plausible to assume that investor recognition of a firm is negatively correlated
with geographical distance. We thus expect the impact of local investors to be particularly
strong for firms which are hardly visible to remote investors:
Hypothesis 2: The negative turnover shock will be more pronounced for those local firms
which are less recognized by non-local investors.
We also explore whether there are differences across investor types, which empirical find-
ings assess to be likely. The aforementioned evidence of limited stock market participation
during holidays appears to hold particularly true for private investors. At the same time,
retail stock ownership tends to be more exposed to local bias than institutional stock hold-
ings (e.g. Grinblatt and Keloharju (2001)). Small firms have been shown to be investment
habitats of retail investors (e.g. Dorn et al. (2008), Kumar and Lee (2006)), whose local
bias is particularly concentrated in these stocks (e.g. Zhu (2003)). Thus, traces of retail
investor behavior in firm-level turnover should be most easily detected in small stocks.
Combined with the observation of Goetzmann and Kumar (2008) that those investors
who trade excessively are particularly locally biased, the rich set of findings suggests:
Hypothesis 3: The negative turnover shock in smaller firms will be disproportionately
caused by individual investors.
7
3. Sample Characteristics
We follow the consensus in the literature on local bias and use a firm’s headquarter as
a proxy for its location. Our initial sample consists of the common stocks of all firms
headquartered in Germany which have been listed on a German stock exchange at some
point between June 13, 1988 and January 15, 2009.3 The lower bound is determined
by the availability of the daily number of shares traded. The upper bound is meant to
maximize the sample size by the inclusion of Epiphany (January 6) in 2009. The data
is then subjected to a three-stage screening process.4 This leaves a final sample of 792
stocks, for which the appendix (available on the web as supplemental material) provides
descriptive statistics at a weekly frequency. The mean (median) firm is in our sample for
556 (515) weeks, has an average market capitalization of 1,148 (123) million Euro, and
has a weekly turnover of 1.42% (0.93%). There is large cross-sectional and considerable
time-series variation in turnover, which again motivates the exploration of local bias as a
potential driver of firm-level trading activity.
Please insert figure 1 and table 1 about here
Table 1 shows summary statistics for event study samples. Several findings highlight
advantages of the German setting. First, both the treatment (holiday) and the control
(non-holiday) groups form large portfolios. Second, their composition does not seem to
differ much. For example, median firms have about the same market capitalization and
comparable average stock returns. Industry concentration, as computed from Herfindahl
3See the appendix, which is available on the web as supplemental material, for an overview of all data sets used in
this study. For the holidays analyzed here, the Frankfurt stock exchange has been open over the whole sample period,
while stock trading at the regional exchanges in Germany started in 2000. This is unlikely to influence our results for three
reasons. First, for all sample stocks, the primary exchange from which Datastream obtains its default prices turns out to
be the Frankfurt stock exchange. Second, inferences remain unchanged if we restrict our analysis to those stocks which are
exclusively traded on the Frankfurt stock exchange. Third, results are robust across time. In particular, they also hold for
the subperiod 2000-2009 (see sections 4.2 and 4.3 for details).
4First, adjusted and unadjusted daily closing prices, market capitalization, book values, the number of daily shares
traded, the number of total shares outstanding, adjustment factors as well as industry membership have to be available
via Datastream. Second, we conduct the tests suggested by Ince and Porter (2006). Third, to assure that our analysis is
not contaminated by very small and illiquid stocks, we exclude securities if their mean market capitalization is less than 10
million Euro or if the 5th percentile of their unadjusted prices is less than 1 Euro. The main results do not change if we use
the sample after step two, which contains 1,071 stocks.
8
indices based on Datastream Level 2 industry classification, is very similar. The appendix
shows that also industry composition appears broadly comparable. Similar findings ap-
ply to the fraction of large firm observations. Third, the time-series properties of local
turnover indices show a remarkably similar behavior, even in the tails of the distribution.
Fourth, an eyeball analysis of figure 1, which shows the geographic distribution of sample
firms, reveals that firm location in Germany tends to be less concentrated than in the
predominantly used US samples (e.g. Ivkovic and Weisbenner (2005)). Fifth, not only
firm-level variables, but also individual investors’ characteristics seem comparable. This
is suggested by calculations based on data of the German SAVE study5, a comprehen-
sive panel survey designed to provide representative information on the financial situation
and relevant socio-psychological traits of German households. For example, households’
propensity to participate in the stock market, investors’ risk taking behavior and eco-
nomic expectations, their financial literacy and use of financial advice, or the influence of
social contacts on financial decision-making is similar in control and treatment groups.
With regard to typical US samples of previous local bias studies, Seasholes and Zhu (2010)
highlight a cross-sectional geographic sampling error, which they argue to potentially lead
to incorrect conclusions. Taken together, the German setting seems to suffer less from
this selection bias. Instead, portfolios are broadly diversified, homogeneous in several
dimensions and thus seem particularly suitable for the following natural experiment.
4. Event Study
4.1 METHODS AND BASELINE RESULTS
In order to quantify the impact of localized trading, one needs to define a measure of
trading activity. We focus on firm turnover as “turnover yields the sharpest empirical
implications and is the most natural measure” (Lo and Wang (2000), p. 12). As turnover
is naturally skewed, we use its natural logarithm in the following calculations. In the
regression setting targeted at testing hypothesis 1, the dependent variable TOi,t is the
daily turnover of firm i on a regional holiday at time t. We consider each year from 1988
5See the appendix for a detailed analysis. For an overview of the SAVE study, see Boersch-Supan et al. (2009).
9
to 2009 in which the holiday falls on a trading day. For All Saints’ Day (Epiphany), this
results in 16 (14) years with a total of 6,485 (5,657) observations.
During regional holidays, market turnover in general tends to be lower. The average daily
turnover of a value-weighted (equal-weighted) turnover index during the whole sample
period is 0.42% (0.20%). On Epiphany, these numbers decrease to 0.36% (0.14%), on All
Saints’ Day to 0.29% (0.13%). However, we are not interested in changes in trading activity
per se, but in potential cross-sectional differences between holiday and non-holiday firms.
Thus, the independent variable of interest is the holiday region dummy Holi,t that equals
one if a firm’s headquarter is located in a holiday region and zero otherwise. The null
hypothesis is that the dummy should not have any significance.
To isolate the holiday effect, it is essential to control for the expected level E[TOi,t]
of turnover. To assure robustness, we rely on two models widely employed in previous
research. Model 1 accounts for firm-specific average turnover in the pre-event period (e.g.
Chae (2005)). In the baseline analysis, the expected firm turnover is calculated as the
natural logarithm of the average turnover over t-20 to t-2. Model 2 controls for both
market-related and firm-specific volume by adopting a “turnover market model”(e.g. Tkac
(1999)). To this end, for t-60 to t-2, turnover for each firm is regressed on a market-
wide, value-weighted turnover index TOm,t. Using the coefficients from the time-series
regression, expected turnover is then given by
E[TOi,t] = αi + βiTOm,t. (1)
As current firm-level turnover might be related to current stock return (e.g. Chordia et al.
(2007)), we include two control variables. Ret+,i,t represents the event day stock return if
positive and zero otherwise.6 Ret−,i,t is defined analogously. This distinction is motivated
by possible asymmetric effects caused by short-selling constraints or the disposition effect,
which have been shown to affect localized trading (e.g. Grinblatt and Keloharju (2001)).
It has also been documented that turnover is influenced by lagged stock returns (e.g.
Statman et al. (2006), Glaser and Weber (2009)). This effect should be captured at least
partly by our measures of expected turnover. To more fully control for recent past returns,
6However, our results do not change if we only include the lagged return or if we do not add return-related control
variables at all. Moreover, as shown in section 4.2, inferences are the same when including interaction terms to allow for a
different impact of returns on holiday firm turnover.
10
we include two analogous variables (Ret+,i,t−1 and Ret−,i,t−1) for the pre-event day return.
The return controls might also be regarded as crude proxies for news or rumors, which
could affect turnover. In section 4.3, we comprehensively test for differences in information
release between holiday and non-holiday firms.
In our basic regression setting, we employ a Fama-MacBeth approach, combined with the
method of Newey and West (1987). We implement the following cross-sectional model in
each year and use the resulting time-series of coefficients to assess their significance:
TOi,t = β0,t + β1,tE[TOi,t] +5∑
k=2
βk,tReturnControlk,i,t + β6,tHoli,t + ϵi,t (2)
Table 2 shows the findings for the Epiphany sample and the All Saints’ Day sample,
respectively. Displayed are results from three regression specifications, which differ in the
dependent variable. The baseline regression uses firm-specific turnover at the day of the
holiday (TOi,t), the others use the day preceding and following the holiday, respectively.
Please insert table 2 about here
The holiday region dummy attains a highly negative coefficient in all specifications. For
both the Epiphany and the All Saints’ Day sample, and for both models of expected
turnover, the coefficient is strongly significant at the one percent level. The upper bounds
of the 95% confidence intervals are all well below zero. Moreover, from an economic
perspective, the effect is quite large: The pure holiday-induced abnormal drop in volume
ranges from roughly 10% to slightly over 20%. Additionally, results are robust across
time: In the Epiphany sample, the holiday region dummy is negative in each year; in the
All Saints’ Day sample, it attains a negative coefficient in about 80% of the observations.
Finally, the holiday effect can, for the most part, only be identified at the day of the holiday
itself. On the day before the holiday, there is no negative shock in trading activity; on the
day after, there is some evidence, which, however, is much weaker than on the date of the
holiday itself.7 In sum, the findings so far support hypothesis 1.
7As unreported findings suggest, the effect on the day after the holiday might at least partly be attributable to the
impact of bridge days as well as the end of Christmas holidays, which varies both across time and states, respectively. This
seems also consistent with the anecdotal evidence given in footnote 2.
11
4.2 ROBUSTNESS CHECKS
The main results from a variety of sensitivity tests are summarized in table 3.
Please insert table 3 about here
Our test specification might be misspecified in the sense that it may lead to a spurious
positive factor loading on the holiday region dummy on average, irrespective of an actual
holiday event. We therefore implement a “placebo treatment”: For each model specifica-
tion and each holiday sample, we randomly select 500 days (excluding the period from
t-1 to t+1, where t denotes the holiday) and, for each of these pseudo events, run the re-
gression as given in Equation (2). Mean and median factor loadings on the holiday region
dummy are given in panel A. In all specifications, they are virtually zero.
There is arguably some element of arbitrariness in the length of the pre-event period in
both models of expected turnover. Therefore, we experimented with intervals from 10 to
100 trading days. Panel B verifies that inferences remain the same.
It might be possible that the importance of the return controls varies between holiday
and non-holiday firms. We thus interact all return variables from the baseline regression
with the regional holiday dummy. It turns out that none of them is significant. Panel C
shows that the importance of the holiday region dummy remains unaffected.
One might be concerned that the results could partially be driven by a disproportionate
number of holiday firms whose stocks are not traded at the event day. Our findings might
then not reflect a broader phenomenon, but rather be attributable to outliers. We thus
repeat the analysis discarding all stocks with zero trading volume. However, as shown in
panel D, this exercise rather strengthens our results.
Panel E shows results when using raw (instead of logarithmized) turnover. In all specifi-
cations, the holiday effect is significant at the 1% level. Moreover, it keeps its economic
significance. For the mean (median) firm the results indicate a pure regional holiday-
induced drop in daily trading volume of roughly 200,000 (more than 20,000) Euro.
12
Residuals of a given firm might be correlated across years, potentially leading to biased
standard errors. We thus follow a suggestion of Petersen (2009) by pooling all firms with
non-zero turnover, adding year dummies and clustering standard errors by firm. As shown
in panel F, findings are robust to this alternative econometric specification.
Legally recognized regional holidays are observed on a state level. Thus, our interpretation
rests on the idea of states being an appropriate classification of the preferred investment
habitats of local investors. While similar concepts have been proven fruitful in US studies
(e.g. Hong et al. (2008), Korniotis and Kumar (2010)), it is clearly only a noisy proxy.
Note, though, that this works against detecting a regional holiday effect: If local investors
tilted their trading towards stocks of local firms irrespective of state borders, then it
would be hard to identify differences in trading activity between two neighboring states.
In an attempt to use a classification scheme with a more pronounced socio-economic
background, we repeat our analysis building on metropolitan areas as defined by the
Conference of Ministers for Spatial Planning.8 Some areas span more than one state,
whereas some states contain more than one metropolitan region. Panel G verifies that the
coefficient is sporadically estimated even marginally more precisely, possibly pointing to
the true impact of localized trading being stronger than reported.
We also study turnover shocks on Corpus Christi as the third legally recognized regional
holiday. It is celebrated in the states of Baden-Wuerttemberg, Bavaria, Hesse, North
Rhine-Westphalia, Rhineland-Palatinate and Saarland, at the Thursday 60 days after
Easter Sunday. Stock market trading on this day started not before 2000, which results in
a total of 5,078 firm-level observations distributed over nine yearly observations. Panel H
verifies that our findings hold also in this case. The shock in turnover is highly significant,
and estimated to be close to 20%. The setting for Corpus Christi is, apart from the shorter
sample period and the fixed day of the week, not conceptually different from Epiphany
and All Saints’ Day. Including all holidays in the remaining tests increases the sample size
and ensures that we consider each regional holiday in Germany for which requirements
on a meaningful event study are met.
8This classification identifies eleven metropolitan regions in which roughly 70% of the German population and 84%
of sample firms are located. http://www.eurometrex.org defines these areas as “larger centres of economic and social life”
containing “core business, cultural and governmental functions”. We only consider areas clearly belonging either to a holiday
or a non-holiday region. This leaves a total of 5,416 (4,350) observations for the All Saints’ Day (Epiphany) sample.
13
However, if our results were representative of a widespread localized trading phenomenon,
then we might also detect similar patterns in related scenarios such as Carnival. While
there is no official holiday, representative surveys reveal that Carnival is prominent in some
(mostly southern and western) regions, but rather unpopular in other (mostly northern
and eastern) areas of Germany.9 Despite the lack of clear-cut separation between affected
and non-affected regions, we run an analogous analysis for Carnival Monday, on which
most parades are held. Panel I provides evidence supportive of our line of reasoning.
4.3 DIFFERENCES IN INFORMATION INTENSITY?
A key concern to a local bias story is that holiday firms might simply release less in-
formation than otherwise comparable non-holiday firms. To the extent that this triggers
rebalancing trades or increased differences of opinion, it might partly explain our findings.
As an intuitive and rather informal first approach to explore the possibility of such an
information effect, we compare the fraction of corporate news released around the holiday.
To this end, we rely on firm-specific news stories published by DGAP, a German news
agency, from January 2000 to January 2009. These news include time-stamped ad hoc
disclosures, by which German firms are forced to publish new value-relevant information
immediately. We manually collect these disclosures for each sample firm. The database
additionally covers a broad range of other news, such as directors’ dealings or business
reports. Since data retrieval is labor intensive, we gather these corporate news for half of
sample firms, which we randomly select. The following test is based on this subsample.
We create a dummy variable that states for each firm and each day whether corporate
news or ad hoc disclosures have been released. Then, for each holiday, and separately for
the holiday and the non-holiday sample, we compute the fraction of all news attributable
to a short window around the holiday (t-1 to t+1). After that, we compute an odds ratio
9We here rely on survey results published in the magazine “Daheim in Deutschland” (by Reader’s Digest), February
2010. Our classification is based on the fraction of individuals stating to actively participate in carnival celebrations. The
areas of Hesse, Rhineland Palatinate, Saarland (roughly 30%), Bavaria (27%), Baden Wurttemberg (25%) and North Rhine-
Westphalia (24%) serve as a treatment group. The remaining regions have participation rates between 10% and 19% and
thus serve as a control group. A related classification scheme based on the relative popularity of carnival clubs leads to
similar results. Data for this analysis is provided by “Bund Deutscher Karneval”, the umbrella organization of several
thousand German carnival clubs.
14
by dividing the percentage obtained for the holiday sample by the percentage obtained for
the non-holiday sample. If holiday firms released temporarily less news, we would expect
values persistently well below one. However, the odds ratios are 1.05 for Epiphany, 1.02 for
All Saint’s Day and 0.84 for Corpus Christi, pointing against a widespread drop in news
release. As later sections of this study reveal that firm size is an important determinant of
the drop in trading activity, we determine whether this might be due to differences in news
release. Specifically, we repeat the analysis separately for large and small stocks, split by
the median of market capitalization at the beginning of the year. Around Epiphany, there
are no marked differences. Around All Saint’s Day, small holiday firms appear to release
relatively more news than large holiday firms. Around Corpus Christi, this picture partly
reverses. In sum, there is no clear pattern.
For deeper insights, we test more rigorously for differences in news arrival from four
further perspectives. Specifically, we study cross-sectional shocks with regard to abnormal
price movements, with regard to the degree of analyst coverage, with regard to investors’
internet search behavior as well as with regard to firms’ media exposure. In the following,
these tests, whose main results are presented in tables 4 and 5, are described in detail.
Please insert table 4 and table 5 about here
Firm-specific news are likely to affect the magnitude of abnormal returns. Firm-specific
information should manifest itself in an increased importance of the idiosyncratic com-
ponent of the firm’s daily stock return. On the other hand, if there is hardly any new
information, then the return should primarily be driven by the stock’s exposure to perva-
sive well-known risk factors. Thus, if there was indeed temporarily less news for the typical
holiday firm, we would expect its absolute abnormal return to be considerably lower than
during some control period on average. For the typical non-holiday firm, however, there
should be no or at least not as much of a difference. A benefit of this approach is that
shock variables can be computed continuously, providing data for each firm on each day.
This overcomes the problem that official news coverage of a given firm may be sporadic,
even though there might be rumors, speculation or private information investors react
on. To formalize the cross-sectional prediction as sketched above, we employ the following
procedure. First, for each firm and each day, we compute the abnormal stock return. By
15
employing both a market model and a Carhart (1997) four factor model, we follow stan-
dard event study methodology; due to very similar findings, only the findings from the
latter model are reported. The four factor model is based on German data and includes
the market, size and value factors in the spirit of Fama and French (1993) and the mo-
mentum factor as constructed in Carhart (1997). The appendix provides more detailed
information about the construction of the factors. Second, for each firm, we compute the
difference between the absolute abnormal return on the day of the holiday (=t) and the
average absolute abnormal return in some control period. We here rely on the period from
t-5 to t+5 (excluding t), but results are not sensitive to this choice. The resulting variable
has the interpretation of an unexpected change in the relative importance of idiosyncratic
stock return factors. Third, for both the holiday and the non-holiday sample, we rank firms
based on this shock variable. We take the cross-sectional median for both samples to get
an estimate of the shock for the typical firm.10 Fourth, we compute the cross-sectional
difference between the median shock for the holiday sample and the median shock for
the non-holiday sample. A news-based explanation of our findings would predict values
significantly below zero, as the shock in the relative importance of firm-specific return fac-
tors for the typical holiday (non-holiday) firm should be more (less) negative. We repeat
the procedure in each year. Finally, a bootstrap approach11 is used to test whether the
average of the resulting time series of differences is statistically distinguishable from zero.
However, panel A of table 4, which reports results for the Epiphany, All Saints’ Day as
well as Corpus Christi sample, shows that this not the case. The only slightly significant
event is on the day of Epiphany, where, from an economic perspective, the resulting re-
turn difference appears small. For all other holidays, differences are very close to zero and
insignificant, implying that in most cases shocks in abnormal returns do not differ much
between holiday and non-holiday firms. Pooling observations does not lead to different
10The appendix provides more details about the distribution of shock variables. It verifies that findings are qualitatively
similar when relying on the mean (instead of the median) of the winsorized cross-section. It also shows that extreme return
events are only slightly more frequent for non-holiday firms.
11The comparison of shock variables results in a holiday-specific time series of differences between holiday and non-
holiday firms. We use this data to simulate 10,000 pseudo time-series of the same length as the original sample by randomly
drawing values with replacement. Averaging values separately for each pseudo time-series yields 10,000 pseudo estimates of
the difference in median shock variables. Finally, we assess whether the value obtained from the averaged original time-series
is reliably negative by computing the fraction of simulated estimates that take on values below zero. For a discussion of
simulations in event studies, see e.g. Lyon et al. (1999).
16
conclusions. Moreover, there are no persistent differences for large and small stocks, again
split by the median of market value.
Our second test is inspired by Da et al. (2011) and based on cross-sectional shocks in search
frequencies for firm names in Google. The application “Google Insights for Search” allows
to construct standardized time-series of terms entered in the internet search engine. Data
is available on a daily basis from January 2004 on. Computing shocks in search volume
might be regarded as a possibility to quantify unexpected changes in revealed (and thus
direct) focus to individual firms, induced by some external stimulus. In this sense, changes
in the query frequency of a firm name12 appear a promising way of capturing shocks in
the arrival of firm-specific news or rumors. For example, Da et al. (2011) report a positive
correlation between search volume shocks and traditional proxies for information release,
such as extreme returns and news stories. The authors show that internet search volume
even often leads alternative measures of news arrival. We thus construct a measure of
unexpected search behavior for each firm based on daily data. It is defined as the difference
between the search frequency during the holiday (=t) minus the average frequency over
t-10 to t-2, divided by the standard deviation in this pre-event period. We then pool
observations and regress the shock variables on a holiday region dummy in addition to
controls for years and industries. We do so for the sample of all firms, of large firms, and
of small firms. Panel B of table 4 shows that all holiday region dummy coefficients are
insignificant, both separately and jointly, pointing against a news-based story.
Our third analysis focuses on the large effort of analysts in collecting, processing and dis-
seminating information (e.g. Womack (1996)). We are interested in whether aggregated
analyst coverage during regional holidays differs from coverage in a nearby benchmark pe-
riod. Specifically, we concentrate on the number of daily analyst recommendations issued,
and determine whether the fraction holiday firms account for is exceptionally low during
12One might be concerned about the use of firm names. They might not be unambiguous and a few of them clearly have
multiple meanings. However, this seems unlikely to drive our main results. First, we study differences between two large
samples with several hundred firm names. Thus, any potential inaccuracies and inconsistencies are likely to cancel out.
Second, we are interested in shocks of search frequencies, i.e. we control for the expected level of queries. Third,“Google
Insights for Search” additionally provides a top search list with the terms most closely related to the original search. In an
attempt to manually cleanse the data, we used that information to exclude those firms that seemed most likely to distort
the analysis. Inferences remained unchanged. The alternative of relying on security identification numbers instead of firm
names turned out to be unproductive as search frequencies tend to be much lower, resulting in many missing values.
17
the holiday. This is what a news-based explanation of our findings would arguably predict.
To test this hypothesis, we match our sample with the I/B/E/S analyst buy/hold/sell-
recommendations database. This results in a total of 51,497 stock recommendations of
196 brokers, which cover more than 80% of the sample firms. For the eleven day pe-
riod centered around the holiday (t-5 to t+5), we then determine which fraction of all
recommendations issued on this day is attributable to holiday firms. The length of this
benchmark period is meant to account for the seasonality in earnings reports, but the
qualitative nature of our findings is robust to alternative control windows. We average
values for t. Values for the benchmark period (excluding t) are pooled to give rise to an
empirical benchmark distribution of relative analyst coverage for holiday firms. Relying
on the percentiles of this distribution, we are able to detect whether analyst informa-
tion transmission for holiday firms exhibits a negative shock. We distinguish between a
value-weighted analysis, in which multiple recommendations made for the same firm on
the same day are considered as multiple observations, and an equal-weighted analysis, in
which we regard such a scenario as a single observation. The latter tends to give more
weight to small firms, which less often receive several recommendations at the same day.
As a sensitivity check, we repeat the analysis now focusing on the review date, i.e. the
most recent date that an estimate is confirmed by an analyst to I/B/E/S as accurate.
Panel C of table 4 shows the fraction of total analyst coverage on the event day. Percentiles
are given in parentheses. A higher percentile indicates that holiday firm recommendations
account for a larger fraction of the total number of recommendations issued. In all speci-
fications, coverage does not seem to decrease for firms located in holiday regions. Judging
from the percentiles of the distribution, the holiday rather appears like an average day
of the benchmark period.13 Moreover, the value-weighted and the equal-weighted analysis
show a similar picture, suggesting there are no marked differences between large and small
firms.
As a final test, we study shocks in media coverage in three leading German daily business
newspapers, which are published nation-wide.14 The comprehensive database, for which
13One might be concerned about noise in the data. Indeed, a similar bootstrapping approach as outlined in footnote 11
reveals that the dispersion of simulated outcomes is quite substantial. However, even the lower bound of the 99% confidence
interval does not touch the 10th percentile of the benchmark distribution, which contradicts an information-based story.
14IVW, a German auditing institution that provides data on the distribution of media products, reports that Sueddeutsche
Zeitung had the second highest circulation among nationwide published daily papers over the period 2000 to 2008. It ranks
18
panel A of table 5 gives more details, is based on daily data from January 1, 2000 on
and comprises Financial Times Deutschland, Handelsblatt and Sueddeutsche Zeitung.
Searching factiva and genios, articles about each firm for each day and in each paper are
manually collected.15 This results in a total of 126,125 news stories covering almost 94% of
our sample firms. Again, we distinguish between a value-weighted and an equal-weighted
analysis. The latter relies on a dummy variable that simply states whether a firm has
received press coverage on a given day. The former individually counts each article. It
thus takes on values greater than one if there are several firm news stories in the same
paper, or if several papers cover the firm. In doing so, it tends to give more weight to
blue chips and big news. For further insights, we additionally split firms into large and
small stocks, as before. Across all specifications, there is considerable variation in daily
media coverage. For instance, on a given day, the fraction of news stories attributable
to firms that didn’t make the news the day before, is 63% (52%) for the equal-weighted
(value-weighted) analysis on average. Focussing on small firms yields even 91% (90%).
Panel B shows results from a test similar to the one used for analyst coverage. We analyze
whether aggregated media coverage for holiday firms is abnormally low around the holiday.
We consider both the event day and the following day, as information becoming public
at t can not be published by newspapers before t+1. To assess statistical significance, we
calculate the percentage of total media coverage attributable to holiday firms for each day
of the year.16 We then analyze the fraction of press coverage around the holiday relative
to the whole empirical distribution, which does not exhibit strong seasonal patterns. The
analysis produces mixed results. Around All Saints’ Day, media coverage for holiday firms
first if one excludes the popular press. Among the daily newspapers with a strong focus on business and economics,
Handelsblatt and Financial Times Deutschland rank first and second. In the fourth quarter of 2008, the three newspapers
had a combined circulation of more than 800,000 copies per day.
15Similarly as in Tetlock et al. (2008), we thereby require the article to mention at least twice the name or security
identification number of the firm. This procedure aims at reducing noise and identifying relevant firm-specific articles.
Coverage for Financial Times Deutschland starts on January 1, 2001.
16We thereby account for the fact that not all newspapers are published at each day of the year: At Corpus Christi,
Handelsblatt and Sueddeutsche Zeitung are not distributed. At Epiphany, Sueddeutsche Zeitung is not published. This is
unlikely to materially influence our analysis. First, for the more important date t+1, all newspapers are available. Findings
are similar as on date t. Second, the results from the equally- and from the value-weighted analysis are similar in general.
This suggests that relevant information is, for the most part, picked up by each of these leading newspapers so that partly
relying on a subset of them does not change the qualitative nature of the results. This line of reasoning is also supported
by the highly significant correlations in daily firm-level media coverage as shown in panel A of table 5.
19
is indeed significantly lower, which, in line with findings from the test on corporate news
releases, appears to be driven by larger firms. However, there is no similar evidence for any
of the other holidays. In fact, press coverage is sometimes even higher than on average.
In the overall picture, results point against a strong general drop in media exposure for
both large and small holiday firms. To gain more insight, we implement a more formal
regression approach. We create the dummy Newsi,t which indicates for each firm i on
each day of the eleven trading days period centered around the regional holiday whether
a news article was published.17 We then pool the observations and run the following probit
regression separately for Epiphany, All Saints’ Day and Corpus Christi:
NEWSi,t = β0 + β1EventDummy + β2HolidayRegionDummy + β3InteractionTerm+ Y earDummies+ ϵi,t
(3)
The event dummy indicates the holiday within the event period. We also run analogous
regressions for the days preceding and following the holiday. Of interest is the interaction
effect between the event dummy and the holiday region dummy. If the volume shock was a
result of systematic cross-sectional differences in press coverage, then it should consistently
attain a significantly negative sign. Panel C of table 5 reports results from the nine probit
regressions. Magnitude and significance of the interaction effect are assessed as suggested
by Ai and Norton (2003). Again, the only significant results are found for the All Saints’
Day sample. Thus, the findings at best sporadically point to differences in information
release picked up by the press.
We finally incorporate additional control variables in our pooled regression approach as
outlined in section 4.2. For data availability reasons, we focus on the period from 2000
to 2009, and add a set of dummies to control for the effect of media coverage and ad
hoc disclosures on any day between t-5 and t+5. Panel D of table 5 reveals that the
regional holiday effect keeps its significance, both from an statistical and an economic
point of view. Modifying the analysis by focussing only on those stocks for which we have
additional information about the release of other corporate news yields similar results.
Taken together, the combined findings from all tests in this section provide the following
17We choose this binary approach to reduce the overcounting of news about the same subject from multiple sources.
However, an analysis focussing on the actual number of news produces very similar results. The eleven day period is largely
representative for the media coverage in the whole year.
20
picture: First, we cannot dismiss the hypothesis of lower information intensity for holiday
firms as there is minor evidence of differences in news release. Their lack of robustness
and small magnitude, however, suggest they are unlikely to fully explain the economically
substantial and pervasive drop in trading volume for holiday firms. The evidence points
against persistent disparities between small and large firms. Second, controlling for po-
tential differences in news arrival to the extent possible, our results remain qualitatively
unchanged. Third, these findings strongly confirm hypothesis 1.
5. Determinants of the Regional Holiday Effect
Firm characteristics What factors drive the cross-sectional heterogeneity in negative
turnover shocks? To answer this question, we first construct a firm-specific measure of ab-
normal turnover, defined as actual (logarithmized) turnover during the holiday minus the
average turnover during t-20 to t-2. For robustness reasons, we then run pooled regressions
separately for each of the three holiday samples as well as for two sample periods.
Hypothesis 2, inspired by the model of Merton (1987), posits that the turnover shock
should be particularly strong if a firm is visible primarily to local investors. Merton argues
that investor recognition is a function of the shadow cost of information, which, in his
model, depends on idiosyncratic risk, relative market size and the completeness of the
shareholder base. We thus use the logarithm of a firm’s market capitalization, as measured
at the end of the preceding year, and a firm’s idiosyncratic risk as independent variables.
Idiosyncratic risk is defined as the standard deviation of the residual obtained by fitting
a Carhart (1997) four factor model (as described in section 4.3) to the daily return time-
series from t-180 to t-6.
Market capitalization is strongly negatively related to the total number of shareholders
(e.g. Grullon et al. (2004)) and positively related to the fraction of local investors (e.g. Zhu
(2003)). Consequently, we expect a smaller drop in volume for larger firms, which implies
a positive coefficient for firm size. Idiosyncratic risk, on the other hand, increases the
shadow cost of information. Local investors are commonly thought to possess (actual or
perceived) informational advantages. Thus, local clienteles should account for a relatively
large proportion in the trading of stocks with high idiosyncratic risk, which should go along
21
with a more pronounced negative volume shock during regional holidays. Consequently,
a negative coefficient is expected.
In addition, we employ with residual media coverage a third proxy, which is orthogonal
to size and available for the years 2001 to 2009. The residual is obtained from yearly
cross-sectional regressions of the number of firm-specific press articles in the previous
year on its lagged average market size, turnover and absolute return as well as on a set
of control variables for industry and DAX30 membership. Press articles are taken from
the comprehensive media coverage database described in section 4.3. Residual coverage is
designed to proxy for the unexpected high or low weight the media attaches to a certain
firm. Given the importance of leading business newspapers in disseminating information
to a broad audience, residual media coverage is an intuitive measure of firm visibility.
Consequently, we expect a positive coefficient.
Previous research and our baseline analysis highlighted the importance of current returns
for current turnover. We thus include the same two return-based variables in the regres-
sion. To control for additional effects induced by medium-term return continuation, we
consider the loading on the momentum factor (WML), obtained from a regression of
stock returns on the Carhart (1997) four factor model. The loadings on the market as
well as value factor (RMRF , HML) are considered as proxies for systematic risk (e.g.
Chordia et al. (2007)). The intercept from this regression (Alpha) is included as it has
been argued to contain a premium related to liquidity or heterogeneous information (e.g.
Lo and Wang (2000)). Moreover, we include a rural dummy for firms located outside a
metropolitan region. The “only game in town effect” (Hong et al. (2008)) suggests a neg-
ative coefficient. Inspired by e.g. Seasholes and Wu (2007), a 52 week high dummy for
stocks whose price has exceeded this bound in the previous week is considered. Finally,
we include a set of industry dummies.
For each holiday, table 6 displays univariate and multivariate results for the whole sample
period. We report coefficients for the subperiod 2001 to 2009 separately. These coefficients
additionally include residual media coverage and controls for the availability of press
articles as well as ad hoc disclosures around the event date (see also section 4.3).
Please insert table 6 about here
22
The findings are broadly consistent with our expectations. Investor recognition seems an
important driver of the turnover shock. All proxies consistently attain the predicted sign
and, with the exception of idiosyncratic risk, are persistently statistically significant. The
effect of market capitalization is clearly the strongest, but residual media coverage has
an incremental effect. The magnitude of the results is also of economic importance: As a
rough estimate, for example, a one standard deviation change in firm size has a similar
impact as a one standard deviation change in stock return. The current absolute return
is highly significant. The dummies for rural firms and the 52 week high attain coefficients
as predicted, but their importance is not robust. The other controls seem to play only a
minor role. In sum, hypothesis 2 can broadly be confirmed. The regional holiday effect is
considerably stronger for firms less visible to non-local investors.
Investor characteristics In this section, we aim at gaining additional insights from the
daily tracking records of roughly 3,000 retail clients of a German online broker from Jan-
uary 1997 to April 2001. Comprehensive information about the sample, such as details
about the construction of portfolio holdings, is given in Glaser and Weber (2009) and
Glaser and Weber (2007). Sample investors account for a total of 316,134 stock trans-
actions, out of which 136,125 take place in 965 German firms. As the latter represent
roughly 50% of all transactions traceable via Datastream, investors seem to exhibit a
strong home bias. Panels A to C of table 7 provide descriptive statistics, which show that
sample investors trade frequently. The mean (median) number of transactions in German
firms is 47 (22), leading to a total sample trading volume of more than 750 million Euro.
Please insert table 7 about here
For the purpose of our analysis, the data set has two advantages. First, the broker does
not offer investment advice. Therefore, trading decisions are not affected by bank recom-
mendations. Second, online broker investor trading on regional holidays is not restricted
in any way. Results suggestive of localized trading might thus be considered conservative
in the sense that other investors might face higher obstacles, such as finding an open bank
office.18 A disadvantage of the sample is that investor location is not provided. Given this
18Note again that this does not have cross-sectional implications for abnormal stock turnover unless local investors’
trading decisions systematically deviate from remote investors’ buys and sells.
23
limitation, exploring to what extent investors exhibit local bias (in addition to home bias),
is not a straightforward exercise. We thus start our analysis with the reasonable assump-
tion that a disproportionate fraction of the broker’s clients live in the region in which the
broker is headquartered. Locally biased investors should then have a strong preference for
firms also located in the affected metropolitan area.19 To test this, we compute a sample
investor preference measure as the difference between a firm’s brokerage weight and its
weight in the market portfolio of German stocks. The firm’s brokerage weight is defined
as the total volume invested in the firm’s stock by the broker’s clients divided by the total
volume the clients invest in all German stocks at the time. We do so at the beginning of
each month and for each firm traded at least once on any day by any sample investor. We
average stock-specific time-series to obtain an average estimate, based on which we sort
firms in one of three portfolios of equal size: “Low preference”, “medium preference” and
“high preference”. Then, for each of these portfolios, we determine the fraction of firms
located in the same metropolitan area as the online broker itself. As the metropolitan area
turns out to be large, there is sufficient level of diversification. Consequently, if sample
investors were not locally biased, we would expect the fraction of firms located near the
online broker to be similar across preference portfolios. However, panel D shows that this
is not what we find. The fraction of the “medium preference” portfolio is standardized
to 1. Therefore, the value 1.29 for the “high preference” portfolio implies that there are
close to 30% more local firms than would be expected on average by chance.
Having verified the existence of at least some local bias, we turn to a test suggested by
hypothesis 1: Holiday trading activity should decrease in local bias. We label each firm
located in the broker’s metropolitan area a “low preference”, “medium preference” or
“high preference” firm. Then, we compute the daily fraction of aggregate sample investor
trading volume that is attributable to each of these portfolios, leading to an empirical
benchmark distribution for portfolio-specific relative trading activity. Similarly as in pre-
vious tests, we determine the percentile of the distribution that is observable during the
day of the regional holiday.20 Hypothesis 1 predicts that these percentiles should decrease
19To sharpen the analysis, we focus on the metropolitan area classification as outlined in section 4.2. Results are similar
when we make use of states instead. Moreover, to mitigate the effect of a few extremely large trades that could materially
affect the analysis, we winsorize investor transactions at the 99.9% level in all following tests.
20To sharpen the analysis, we focus on the holiday that most clearly separates the broker’s metropolitan area as a holiday
region from as many other metropolitan areas as possible.
24
in local investors’ preference - firms with a high degree of local bias should exhibit a more
pronounced shock in relative trading volume. Panel E shows that this is indeed what
we find. The “high preference portfolio” temporarily exhibits the lowest trading activity,
no matter if one focuses on the total Euro volume traded, the number of transactions
conducted or the number of investors trading.
We now turn to hypothesis 3, which posits that the turnover drop in small stocks is dis-
proportionately caused by private investors. To this end, we aggregate data and conduct
tests based on shocks in a measure called Ratioi,t. For holiday i, it is computed as the
overall fraction of daily “holiday firm trading” by online broker investors divided by the
fraction of daily “holiday firm trading” by the whole market. The rationale is as follows:
As the daily trading volume of the investor sample is positively correlated (0.39) with the
daily market trading volume for these firms, it appears justified to use market volume
as a benchmark. By focussing on shocks of Ratioi,t, one mitigates the problem of lacking
information on investor location, as the expected level of trading in each group of stocks
is automatically accounted for. To identify shocks, we control for the autoregressive prop-
erties of Ratioi,t by employing AR(p)-processes similar to Connolly and Stivers (2003).
Shocks are defined as the residual ϵi,t from the following regression:
Ratioi,t = βi,0 +p∑
k=1
βi,kRatioi,t−k + ϵi,t (4)
P denotes the maximum lag, up to which each estimated coefficient on each lagged term
of Ratioi,t is individually significant, and takes on values between two and five for the
specifications described below. ϵi,t can thus be interpreted as unexpected daily changes in
holiday firm trading of retail investors as compared to the whole market.
To test hypothesis 3, we compute Ratioi,t separately for the whole sample as well as for
small and large stocks, split by the median of market capitalization at the beginning
of the year. We do this for each of the three holidays. We then determine the most
suitable AR(p)-process for each of the nine specifications and run the regression as given
in Equation (4). This results in nine shock time series. Finally, we apply these to the seven
holiday observations that take place on a trading day during our retail investor sample
period: Epiphany is celebrated four times, All Saints’ Day twice and Corpus Christi once.
Panel F of table 7 reports the percentiles of the shock variables for each stock sample (all,
25
large, small). The results for large stocks and for the whole sample appear like random
draws from the distribution. In other words, there are no systematic differences between
individual investors and the overall market. However, focussing explicitly on smaller firms,
a clear pattern emerges: Online broker investors’ trading activity consistently exhibits
negative shocks at the day of the holiday when benchmarked against the whole market.
The value of the shock variable is well below its median for every single observation.
Assuming independence, the likelihood of observing this result by chance is below 1%. In
other words, the findings are consistent with hypothesis 3.
6. Conclusion
We run a series of natural experiments which collectively suggest that local bias leaves
discernible traces in the cross-section of firm-level trading activity. The German setting
allows us to compare abnormal turnover in several treatment groups, i.e. hundreds of
firms in holiday regions, with turnover in control groups, i.e. in many ways very similar
firms in non-holiday regions. Ceteris paribus, firms in holiday regions are remarkably
less traded. This finding is mostly confined to the day of the holiday itself, statistically
significant, economically meaningful, robust, and does not appear to be completely driven
by differences in information release. Instead, consistent with a local bias explanation and
the model of Merton (1987), it is particularly strong for firms less recognized by non-local
investors. Moreover, in line with predictions of previous research, the turnover shock in
smaller stocks seems to be disproportionately caused by individual investors.
The basic message of this study is a simple one: Local investor clienteles are strong and
pervasive enough to generate frictions segmenting the stock market along a geographical
line. Our analysis also contributes to research on determinants of firm-level trading volume
by establishing cross-sectional regularities related to firm location, firm visibility, and
investor clienteles. Moreover, by uncovering a link between the potentially powerful role
of local investors, investor distraction, and the cross-section of firm turnover, we might
provide a new fruitful starting point for the emerging research on the joint dynamics of
investor attention, trading volume, and price discovery.
26
References
Ai, C., and E. C. Norton, 2003, “Interaction terms in logit and probit models,” Economic
Letters, 80, 123–129.
Baik, B., J.-K. Kang, and J.-M. Kim, 2010, “Local institutional investors, information
asymmetries, and equity returns,” Journal of Financial Economics, 97, 81–106.
Bailey, W., A. Kumar, and D. Ng, 2008, “Foreign investments of U.S. individual investors:
Causes and consequences,” Management Science, 54, 443–459.
Barberis, N., A. Shleifer, and J. Wurgler, 2005, “Comovement,” Journal of Financial
Economics, 75, 283–317.
Bodnaruk, A., 2009, “Proximity always matters: Local bias when the set of local companies
changes,” Review of Finance, 13, 629–656.
Boersch-Supan, A., M. Coppola, L. Essig, A. Eymann, and D. Schunk, 2009, “The German
SAVE study - Design and results,” MEA Study No. 6.
Brown, J. R., Z. Ivkovic, P. A. Smith, and S. Weisbenner, 2008, “Neighbors matter: Causal
community effects and stock market participation,” Journal of Finance, 63, 1509–1531.
Carhart, M. M., 1997, “On persistence in mutual fund performance,” Journal of Finance,
52, 57–82.
Chae, J., 2005, “Trading volume, information asymmetry, and timing information,” Jour-
nal of Finance, 60, 413–442.
Chordia, T., S.-W. Huh, and A. Subrahmanyam, 2007, “The cross-section of expected
trading activity,” Review of Financial Studies, 20, 709–740.
Connolly, R., and C. Stivers, 2003, “Momentum and reversals in equity-index returns
during periods of abnormal turnover and return dispersion,” Journal of Finance, 58,
1521–1556.
Coval, J. D., and T. J. Moskowitz, 1999, “Home bias at home: Local equity preference in
domestic portfolios,” Journal of Finance, 54, 2045–2073.
, 2001, “The geography of investment: Informed trading and asset prices,” Journal
of Political Economy, 109, 811–841.
27
Da, Z., J. Engelberg, and P. Gao, 2011, “In search of attention,” forthcoming Journal of
Finance.
DellaVigna, S., and J. M. Pollet, 2009, “Investor inattention and Friday earnings an-
nouncements,” Journal of Finance, 64, 709–749.
Dorn, D., and G. Huberman, 2005, “Talk and action: What individual investors say and
what they do,” Review of Finance, 9, 437–481.
Dorn, D., G. Huberman, and P. Sengmueller, 2008, “Correlated trading and returns,”
Journal of Finance, 63, 885–920.
Fama, E. F., and K. R. French, 1993, “Common risk factors in the returns on stocks and
bonds,” Journal of Financial Economics, 33, 3–56.
Feng, L., and M. S. Seasholes, 2004, “Correlated trading and location,” Journal of Fi-
nance, 59, 2117–2144.
Frieder, L., and A. Subrahmanyam, 2004, “On non-secular regularities in trading activity
and stock returns,” Financial Analysts Journal, 60, 29–34.
Gao, X., and T.-C. Lin, 2010, “Do behavioral needs influence the trading activity of in-
dividual investors? Evidence from repeated natural experiments,” Unpublished working
paper, University of Hong Kong.
Glaser, M., and M. Weber, 2007, “Overconfidence and trading volume,” Geneva Risk and
Insurance Review, 32, 1–36.
, 2009, “Which past returns affect trading volume?,” Journal of Financial Markets,
12, 1–31.
Goetzmann, W. N., and A. Kumar, 2008, “Equity portfolio diversification,” Review of
Finance, 12, 433–463.
Grinblatt, M., and M. Keloharju, 2001, “How distance, language, and culture influence
stockholdings and trades,” Journal of Finance, 56, 1053–1073.
Grullon, G., G. Kanatas, and J. P. Weston, 2004, “Advertising, breadth of ownership,
and liquidity,” Review of Financial Studies, 17, 439–461.
28
Hau, H., 2001, “Location matters: An examination of trading profits,” Journal of Finance,
56, 1959–1983.
Hirshleifer, D., S. S. Lim, and S. H. Teoh, 2009, “Driven to distraction: Extraneous events
and underreaction to earnings news,” Journal of Finance, 64, 2289–2325.
Hong, H., J. D. Kubik, and J. C. Stein, 2004, “Social interaction and stock-market par-
ticipation,” Journal of Finance, 59, 137–163.
, 2005, “Thy neighbor’s portfolio: Word-of-mouth effects in the holdings and trades
of money managers,” Journal of Finance, 60, 2801–2824.
, 2008, “The only game in town: Stock-price consequences of local bias,” Journal
of Financial Economics, 90, 20–37.
Hong, H., and J. Yu, 2009, “Gone fishin’: Seasonality in trading activity and asset prices,”
Journal of Financial Markets, 12, 672–702.
Hou, K., L. Peng, and W. Xiong, 2009, “A tale of two anomalies: The implications of
investor attention for price and earnings momentum,” Unpublished working paper, Ohio
State University, City University of New York, Princeton University.
Huberman, G., 2001, “Familiarity breeds investment,” Review of Financial Studies, 14,
659–680.
Ince, O. S., and R. B. Porter, 2006, “Individual equity return data from Thomson Datas-
tream: Handle with care!,” Journal of Financial Research, 29, 463–479.
Ivkovic, Z., and S. Weisbenner, 2005, “Local does as local is: Information content of the
geography of individual investors’ common stock investments,” Journal of Finance, 60,
267–306.
, 2007, “Information diffusion effects in individual investors’ common stock pur-
chases: Covet thy neighbors’ investment choices,” Review of Financial Studies, 20, 1327–
1357.
Korniotis, G. M., and A. Kumar, 2010, “State-level business cycles and local return pre-
dictability,” Unpublished working paper, University of Miami.
Kumar, A., and C. M. Lee, 2006, “Retail investor sentiment and return comovements,”
Journal of Finance, 61, 2451–2486.
29
Lo, A. W., and J. Wang, 2000, “Trading volume: Definitions, data analysis, and implica-
tions of portfolio theory,” Review of Financial Studies, 13, 257–300.
Loughran, T., and P. Schultz, 2004, “Weather, stock returns, and the impact of localized
trading behavior,” Journal of Financial and Quantitative Analysis, 39, 343–364.
, 2005, “Liquidity: Urban versus rural firms,” Journal of Financial Economics, 78,
341–374.
Lyon, J. D., B. M. Barber, and C.-L. Tsai, 1999, “Improved methods for tests of long-run
abnormal stock returns,” Journal of Finance, 54, 165–201.
Massa, M., and A. Simonov, 2006, “Hedging, familiarity and portfolio choice,” Review of
Financial Studies, 19, 633–685.
Merton, R. C., 1987, “A simple model of capital market equilibrium with incomplete
information,” Journal of Finance, 42, 483–510.
Newey, W. K., and K. D. West, 1987, “A simple, positive semi-definite, heteroskedasticity
and autocorrelation consistent covariance matrix,” Econometrica, 55, 703–708.
Petersen, M. A., 2009, “Estimating standard errors in finance panel data sets: Comparing
approaches,” Review of Financial Studies, 22, 435–480.
Pirinsky, C., and Q. Wang, 2006, “Does corporate headquarters location matter for stock
returns?,” Journal of Finance, 61, 1991–2015.
Seasholes, M. S., and G. Wu, 2007, “Predictable behavior, profits, and attention,” Journal
of Empirical Finance, 14, 590–610.
Seasholes, M. S., and N. Zhu, 2010, “Individual investors and local bias,” Journal of
Finance, 65, 1987–2011.
Shive, S. A., 2010, “An epidemic model of investor behavior,” Journal of Quantitative
and Financial Analysis, 45, 169–198.
, 2011, “Local investors, price discovery, and market efficiency,” Unpublished work-
ing Paper, University of Notre Dame.
Statman, M., S. Thorley, and K. Vorkink, 2006, “Investor overconfidence and trading
volume,” Review of Financial Studies, 19, 1531–1565.
30
Tetlock, P. C., M. Saar-Tsechansky, and S. Macskassy, 2008, “More than words: Quanti-
fying language to measure firms’ fundamentals,” Journal of Finance, 63, 1437–1467.
Tkac, P. A., 1999, “A trading volume benchmark: Theory and evidence,” Journal of
Financial and Quantitative Analysis, 34, 89–114.
Womack, K. L., 1996, “Do brokerage analysts’ recommendations have investment value?,”
Journal of Finance, 51, 137–167.
Zhu, N., 2003, “The local bias of individual investors,” Unpublished working paper, Yale
University.
31
Figure 1: Geographic Distribution of Firm Headquarters and of Regional Holidays across Germany
This figure shows the location of firm headquarters across Germany. Headquarters are represented by black
circles; additional clusters of headquarters (with more than 20 firms) in a given city are represented by red
numbers. These correspond, from west to east, to the cities of Duesseldorf (28 firms), Cologne (36 firms),
Frankfurt (40 firms), Stuttgart (21 firms), Hamburg (57 firms), Munich (70 firms) and Berlin (48 firms).
Moreover, the figure exemplarily illustrates the geographic distribution of regional holidays across Germany.
Shown is the example of Epiphany, which is legally recognized only in the grey-shaded states of Baden-
Wuerttemberg (118 firms), Bavaria (180 firms), and Saxony-Anhalt (3 firms).
48
57
21
70
40
36
28
32
Tab
le1:
DescriptiveStatisticsof
EventStudySam
ples
This
table
provides
summary
statisticsofev
entstudysamplesforEpiphany(P
anel
A)andAllSaints’Day(P
anel
B).
Holiday(N
on−
Holiday)den
otesth
esample
of
firm
swhose
hea
dquarter
is(n
ot)
loca
tedin
aregionwhereth
eresp
ectiveholidayis
legallyreco
gnized
.M
edianmaca
pandmed
iandailyreturnreferto
thecross-sectional
med
ianoftimeseries
averages
ofsample
firm
s’market
capitaliza
tion(inmillionEuro)anddailyretu
rns,
resp
ectively.
IndustryConcentration
den
otesth
eHerfindahl
index
,co
mputedasth
esu
mofsquaredindustry
groupweights
(Datastream
Lev
el2industry
classifica
tion,10groups).Twoversionsare
shown.In
theeq
ual-weighted
version(E
W),
industry
weights
are
determined
byth
epercentageofsample
industry
groupfirm
s.In
thevalue-weightedversion(V
W),
industry
weights
are
determined
byth
erelativemarket
capitaliza
tionofindustry
groupobservations,
thereb
yex
cludingDAX30firm
s.Table
2ofth
eappen
dix
provides
more
detailed
inform
ationabout
industry
concentrationandindustry
composition.DAX
(DAX/M
DAX)refers
toth
efractionofsample
observationsattributable
toDAX30(D
AX30andMDAX)firm
s
inth
eperiod1988-2009(1994-2009).
TheMDAX
comprises50largeGermanco
mpaniesfrom
traditionalsectors
rankingim
med
iately
after
the30DAX
firm
s.
Panel
A:EpiphanySample
Region
Number
offirm
sMed
ianMacap
Dailymed
ianreturn
Industry
ConcentrationEW
(VW
)DAX
(DAX
+MDAX)
Holiday
301
1,037
-0.01%
0.15(0.16)
6.1%
(14.2%)
Non-H
oliday
491
977
-0.01%
0.15(0.16)
5.0%
(15.0%)
Tim
e-series
properties
ofequallyweightedturnov
erindices
basedondailydata
Region
Mean
SD
5th
Percentile
95th
Percentile
Correlation(H
oliday,Non-H
oliday
)
Holiday
0.21%
0.13%
0.06%
0.45%
0.86
Non-H
oliday
0.20%
0.12%
0.07%
0.43%
Panel
B:AllSaints’Day
sample
Region
Number
offirm
sMed
ianMacap
Dailymed
ianreturn
Industry
ConcentrationEW
(VW
)DAX
(DAX
+MDAX)
Holiday
509
1,155
-0.01%
0.15(0.17)
5.9%
(14.9%)
Non-H
oliday
283
704
-0.02%
0.16(0.18)
4.3%
(14.2%)
Tim
e-series
properties
ofequallyweightedturnov
erindices
basedondailydata
Region
Mean
SD
5th
Percentile
95th
Percentile
Correlation(H
oliday,Non-H
oliday
)
Holiday
0.20%
0.12%
0.06%
0.44%
0.83
Non-H
oliday
0.21%
0.13%
0.07%
0.46%
33
Tab
le2:
Firm
Turnover
AroundEpiphan
yan
dAllSaints’Day
:YearlyFama-MacB
ethRegressions
This
table
summarizesresu
ltsfrom
variousregressions.
Thedep
enden
tvariable
isth
enatu
rallogarith
moffirm
turn
over
onth
edayof(t),
priorto
(t−
1)orfollowing
(t+1)th
eholiday.
InPanel
AandC,ex
pectedturnover
ismea
suredasth
eaveragefirm
turn
over
over
t-20to
t-2.In
Panel
BandD,amarket
model
oftu
rnover
calibrated
from
t-60to
t-2is
employed
instea
d.Ret
+,t
(Ret
+,t−1)den
ote
thestock
retu
rnatt(t-1)if
positiveand
zero
oth
erwise.
Ret
−,t
(Ret
−,t−1)are
defi
ned
analogously.
T-statistics(inparenth
eses)are
computedwithNew
ey/W
est(1987)-adjusted
standard
errors.Statisticalsignifica
nce
atth
eten,fiveandone-percentlevel
isindicatedby
*,**,and***,resp
ectively.
Thereported
R2is
calculatedasth
eaverageofadjusted
R2.Pred
ictedsignden
otesth
efractionofaneg
ativeholidaydummyco
efficien
tatt.
PanelA:Firm
SpecificExpected
Turn
over(N
=5,657)Aro
und
Epiphany
DependentVariable
Constant
Expected
Turn
over
Ret +
,tRet −
,tRet +
,t−
1Ret −
,t−
1Reg
ionalholiday
dummy
R2
Firm
turn
overatt-1
-1.49***(-6.13)
0.94***(3
3.99)
11.00***(8
.49)
8.97**(2
.71)
3.69(1
.25)
-1.65(-0.42)
-0.05
(-1.21)
0.71
Firm
turn
overatt
-1.91***(-6.77)
0.91***(2
6.66)
19.25***(4
.28)
18.90***(3
.24)
4.63*(1
.84)
-3.45(-1.53)
-0.23***
(-5.59)
0.69
Firm
turn
overatt+
1-1.50***(-6.32)
0.94***(3
7.97)
16.54***(6
.09)
13.86***(6
.01)
5.38*(2
.02)
4.70(1
.34)
-0.10**
(-2.40)
0.71
Predicted
sign
14/14
95%
conf.
interval
-0.32/-0.14
PanelB:M
ark
etM
odelExpected
Turn
over(N
=5,657)Aro
und
Epiphany
DependentVariable
Constant
Expected
Turn
over
Ret +
,tRet −
,tRet +
,t−
1Ret −
,t−
1Reg
ionalholiday
dummy
R2
Firm
turn
overatt-1
-0.64***(-4.02)
0.96***(8
3.87)
12.46***(9
.07)
11.64***(3
.52)
1.74(0
.82)
1.54(0
.42)
-0.00
(-0.15)
0.79
Firm
turn
overatt
-0.97***(-5.60)
0.94***(7
0.28)
19.41***(4
.85)
16.77***(3
.14)
5.20**(2
.63)
-0.83(-0.72)
-0.18***
(-9.33)
0.79
Firm
turn
overatt+
1-0.58***(-3.82)
0.96***(8
6.28)
16.59***(6
.18)
14.46***(5
.78)
5.86***(3
.12)
3.09(0
.97)
-0.05*
(-2.11)
0.80
Predicted
sign
14/14
95%
conf.
interval
-0.22/-0.14
PanelC:Firm
SpecificExpected
Volume(N
=6,485)Aro
und
All
Saints’Day
DependentVariable
Constant
Expected
Turn
over
Ret +
,tRet −
,tRet +
,t−
1Ret −
,t−
1HolidayReg
ion
Dummy
R2
Firm
turn
overatt-1
-1.44***(-8.84)
0.94***(4
9.02)
15.05***(5
.57)
13.13***(5
.01)
2.81(1
.46)
0.90(0
.40)
-0.00
(-0.11)
0.76
Firm
turn
overatt
-2.64***(-8.73)
0.85***(2
3.76)
22.21***(5
.36)
20.09***(5
.01)
5.63***(3
.85)
-3.87(-1.64)
-0.13***
(-4.03)
0.70
Firm
turn
overatt+
1-1.60***(-8.02)
0.92***(3
6.16)
13.74***(4
.88)
15.76***(4
.86)
8.98***(3
.04)
6.41***(4
.46)
-0.06*
(-1.80)
0.75
Predicted
sign
13/16
95%
conf.
interval
-0.20/-0.06
PanelD:M
ark
etM
odelExpected
Volume(N
=6,485)Aro
und
All
Saints’Day
DependentVariable
Constant
Expected
Turn
over
Ret +
,tRet −
,tRet +
,t−
1Ret −
,t−
1HolidayReg
ion
Dummy
R2
Firm
turn
overatt-1
-0.58***(-4.69)
0.96***(6
5.99)
14.20***(4
.33)
15.53***(4
.90)
6.21***(3
.42)
0.87(0
.40)
-0.04
(-0.76)
0.80
Firm
turn
overatt
-1.72***(-9.86)
0.88***(5
9.10)
22.48***(6
.10)
20.02***(5
.80)
5.11***(3
.89)
-1.11(-0.62)
-0.16***
(-4.87)
0.77
Firm
turn
overatt+
1-0.62***(-5.64)
0.96***(6
5.00)
14.19***(4
.01)
14.38***(5
.06)
6.80***(3
.43)
6.41***(3
.82)
-0.10**
(-2.64)
0.80
Predicted
sign
13/16
95%
conf.
interval
-0.23/-0.09
34
Table 3: Robustness Checks
This table displays the coefficient in front of the holiday dummy obtained from various regressions to test for
the robustness of our baseline results. T-statistics are reported in parentheses. Statistical significance at the
ten, five and one-percent level is indicated by *, **, and ***, respectively.
Epiphany All Saints’ Day
Panel A: Mean and Median Factor Loadings on the Regional Holiday Dummy from Placebo Treatments
Firm specific expected turnover Mean: -0.001, Median: -0.004 Mean: 0.005, Median: 0.001
Market model expected turnover Mean:0.007, Median: 0.009 Mean: 0.005, Median: 0.006
Panel B: Alternative Pre-event Periods
Firm specific expected turnover in t-10 to t-2 -0.24*** (-5.54) -0.13*** (-3.45)
Firm specific expected turnover in t-40 to t-2 -0.21*** (-5.02) -0.14*** (-3.78)
Firm specific expected turnover in t-40 to t-11 -0.20*** (-5.28) -0.15*** (-4.14)
Market model expected turnover in t-100 to t-2 -0.17*** (-7.20) -0.17*** (-3.70)
Market model expected turnover in t-40 to t-2 -0.19*** (-9.04) -0.16*** (-5.41)
Market model expected turnover in t-60 to t-11 -0.17*** (-9.12) -0.16*** (-4.62)
Panel C: Interacting Return Variables with Holiday Dummies
Firm specific expected turnover -0.29*** (-4.29) -0.17*** (-2.82)
Market model expected turnover -0.22*** (-5.69) -0.21*** (-3.54)
Panel D: Omitting Stocks With Zero Trading Volume on Event Day
Firm specific expected turnover -0.24*** (-11.48) -0.15*** (-4.84)
Market model expected turnover -0.20*** (-15.10) -0.19*** (-6.23)
Panel E: Using Ordinary Turnover
Firm specific expected turnover -0.02%*** (-3.80) -0.02%*** (-3.10)
Market model expected turnover -0.02%*** (-3.31) -0.02%*** (-3.82)
Panel F: Pooled Regression With Year Dummies and Standard Errors Clustered by Firm
Firm specific expected turnover -0.23*** (-4.73) -0.13** (-2.51)
Market model expected turnover -0.19*** (-4.13) -0.16*** (-2.86)
Panel G: Analysis Based on Metropolitan Areas
Firm specific expected turnover -0.22*** (-9.46) -0.13*** (-4.13)
Market model expected turnover -0.18*** (-5.90) -0.17*** (-4.77)
Panel H: Regional Holiday Effects on Corpus Christi (Since 2000, Econometric Approach as in Panel F)
Firm specific expected turnover Market model expected turnover
-0.20*** (-2.67) -0.20*** (-3.61)
Panel I: Carnival Monday
Firm specific expected turnover Market model expected turnover
-0.05** (-2.38) -0.06** (-2.74)
35
Table 4: Tests for Cross-Sectional Differences in News Arrival
This table summarizes results from various tests aimed at detecting potential cross-sectional differences in news
arrival between holiday and non-holiday firms at the day of the holiday (=t). Large firms (Small firms) refer
to stocks with a market value larger (smaller) than the median stock, measured at the beginning of the year.
Statistical significance at the ten, five and one percent level is indicated by *, **, and ***, respectively. Panel
A reports differences in shocks in absolute abnormal returns. To this end, daily absolute abnormal returns for
each firm during t-5 to t+5, as obtained from a German version of the Carhart (1997) four factor model, are
computed. Factor loadings are estimated from time-series regressions from t-66 to t-6. For both the holiday
and the non-holiday sample, firm-specific shocks are computed as the absolute abnormal return at t minus
the average absolute abnormal return in t-5 to t+5 (excluding t). The table reports the difference between the
median shock value for the holiday sample and the median shock value for the non-holiday sample, averaged
across years. Statistical significance is assessed by bootstrapping as described in footnote 11. Panel B reports
the coefficient in front of the regional holiday dummy as obtained from pooled regressions of daily firm-specific
abnormal search volume in Google on dummies for regional holidays, years and industry groups. Abnormal
search volume is computed as the difference between the search volume at t and the average search volume
in t-10 to t-2, divided by the standard deviation of search volume in this pre-event period. T-statistics are
reported in parentheses. The last column reports p-values as obtained from an F-test of joint significance of
all three holiday dummies. Panel C shows the average fraction of total analyst recommendations and reviews
attributable to holiday firms at t. Only recommendations and reviews issued (not outstanding) on a given day
are considered. In a similar way, the fraction holiday firms account for is also computed for every other day in
t-5 to t+5. These values are pooled to construct an empirical benchmark distribution of analyst coverage in
a nearby period. Values in parentheses represent the percentiles of this distribution as achieved at t. A higher
percentile indicates that holiday firm recommendations account for a larger fraction of the total number of
recommendations. In the value-weighted (equal-weighted) analysis, multiple recommendations of the same
firm are considered as multiple observations (single observation).
Panel A: Differences in Shocks in Absolute Abnormal Returns
Dependent Variable Epiphany All Saints’ Day Corpus Christi Pooled
Difference in shock variable: All firms -0.05%* -0.02% -0.02% -0.03%
Difference in shock variable: Large firms -0.07%* -0.01% 0.01% -0.03%
Difference in shock variable: Small firms -0.02% -0.05% -0.03% -0.03%
Panel B: Abnormal Search Frequencies for Firm Names in Google
Dependent Variable Epiphany All Saints’ Day Corpus Christi P-value joint sign.
Shocks in online search queries: All firms -0.13 (-0.61) -0.15 (-1.40) -0.19 (-1.42) 0.17
Shocks in online search queries: Large firms -0.26 (-0.52) -0.33 (-1.02) -0.20 (-0.40) 0.19
Shocks in online search queries: Small firms -0.12 (-0.58) -0.16 (-1.55) -0.08 (-1.14) 0.16
Panel C: Fraction of Holiday Firm Analysts Recommendations
Dependent Variable Epiphany All Saints’ Day Corpus Christi
Value-weighted fraction of recommendations 40.34% (64) 68.57% (59) 81.51% (40)
Equally-weighted fraction of recommendations 40.37% (58) 69.90% (62) 82.57% (43)
Value-weighted fraction of reviews 41.67% (66) 63.64% (40) 87.87% (54)
Equally-weighted fraction of reviews 43.08% (72) 63.84% (44) 89.23% (54)
36
Table
5:Med
iaCoverag
ean
dtheRegional
Holiday
Effect
Panel
Bsh
owsth
eaveragefractionofnew
sstories
forholidayfirm
sont-1,t(theholiday)andt+
1.This
fractionis
alsoco
mputedforev
eryoth
erdayofth
eyea
r,giving
rise
toaben
chmark
distribution.Percentilesobtained
fort-1,tand
t+1are
reported
inparenth
eses.A
higher
valueindicatesth
atholidayfirm
saccountforalarger
fractionoftotalcoverage.
Largefirms(S
mallfirms)
referto
stock
swithamarket
valuelarger
(smaller)th
anth
emed
ianstock
.Panel
Creportsth
einteractioneff
ect,
computedasin
AiandNorton(2003),
obtained
from
probit
regressionsofpress
coverageonth
eregionalholidaydummy,
theev
entdate,holidaydummy*ev
entdate,and
yea
rdummies.
Panel
Dsh
owsth
eco
efficien
tsforth
eholidaydummy,
asobtained
byth
ebaselinepooledregression(see
section4.2)plusasetofdummiesfordailypress
coverageandadhocdisclosu
reswithin
t-5to
t+5.Statisticalsignifica
nce
atth
eten,fiveandonepercentlevel
isindicatedby*,**,and***,resp
ectively.
PanelA:DescriptiveSta
tistics(J
anuary
1,2000-January
15,2009)
Newsp
aper
Tota
lnumberofnewsstories
Covera
ge
PairwiseFirm-L
evelCorrelation
inM
edia
Covera
ge
Handelsblatt
61,956
90.34%
Handelsblatt
SZ
FTD
SZ
34,497
79.42%
Handelsblatt
10.23***
0.20***
FTD
29,672
73.44%
SZ
0.23***
10.19***
All
126,125
93.77%
FTD
0.20***
0.19***
1
PanelB:Fra
ction
ofNewsSto
riesaboutHolidayFirmsAro
und
theHoliday(=
t,Percentilesin
Parenth
eses)
Holiday
value-w
eighted
equal-weighted
tt+
1t
t+1
Epiphany:All
firm
s30.19%
(88)
22.91%
(14)
32.41%
(89)
27.05%
(53)
All
Saints’Day:All
firm
s55.83%
(3)*
*55.85%
(3)*
*59.33%
(4)*
*59.28%
(3)*
*
Corp
usChristi:
All
firm
s81.68(3
9)
82.63%
(46)
81.21(3
4)
81.32%
(34)
Epiphany:Larg
efirm
s30.02%
(89)
22.11%
(14)
31.55%
(87)
26.12%
(45)
All
Saints’Day:Larg
efirm
s55.33%
(4)*
*55.64%
(4)*
*58.56%
(3)*
*59.78%
(4)*
*
Corp
usChristi:
Larg
efirm
s82.35(3
7)
82.48%
(37)
82.01(3
5)
81.52%
(34)
Epiphany:Small
firm
s33.33%
(55)
29.41%
(44)
33.33%
(55)
29.41%
(44)
All
Saints’Day:Small
firm
s71.43%
(53)
57.14%
(20)
64.69%
(43)
57.14%
(20)
Corp
usChristi:
Small
firm
s66.67(3
8)
82.98%
(72)
66.67(3
8)
82.98%
(72)
PanelC:M
arg
inalEffects
ofPro
bit
RegressionsAro
und
theHoliday(=
t,z-valuesin
Parenth
eses)
Holiday
t-1
tt+
1
Epiphany
-0.41%
(-0.69)
0.70%
(1.16)
-0.03%
(-0.05)
All
Saints’Day
-0.01%
(-0.02)
-1.17%**(-2.15)
-1.16%*(-1.86)
Corp
usChristi
0.35%
(0.71)
-0.33%
(-0.60)
-0.80%
(-1.35)
PanelD:Pooled
Regression
(2000-2009)with
Controls
forM
edia
Covera
geand
Ad
HocDisclosu
res(t-sta
tisticsin
Parenth
eses)
Epiphany
All
Saints’Day
Corp
usChristi
Firm-specificexpected
turn
over
-0.20***(-3.86)
-0.11**(-1.98)
-0.16**(-2.16)
Mark
etmodelexpected
turn
over
-0.19***(-3.58)
-0.12**(-2.22)
-0.17**(-2.23)
37
Table
6:Determinan
tsof
theRegional
Holiday
Effect
This
table
summarizesth
emain
resu
ltsobtained
from
pooled
regressionsoffirm
-specificabnorm
altu
rnover
on
thedate
ofa
regionalholiday
(=t)
on
anumber
of
explainingvariables.
Only
firm
swhose
hea
dquarter
isloca
tedwithin
theregionwhereth
eholidayis
legallyreco
gnized
are
included
.Theindep
enden
tvariablesinclude
retu
rnco
ntrols
(see
table
2foradetailed
description);
thelogoflagged
firm
market
capitaliza
tion;idiosyncraticrisk;residualmed
iacoverage;
theloadingsonth
emarket,
valueandmomen
tum
factor(R
MRF,HML,W
ML,resp
ectively);
theintercep
tfrom
thatregression(A
lpha);
aru
raldummy;a52weekhighdummy;co
ntrols
formed
ia
coverageandadhocdisclosu
resaroundth
eev
ent;
asetof10industry
dummiesobtained
from
theDatastream
level
2industry
classifica
tion;yea
rdummies.
Standard
errors
are
clustered
byfirm
.Statisticalsignifica
nce
atth
eten,fiveandone-percentlevel
isindicatedby*,**,and***,resp
ectively.
PanelA:Determ
inants
ofth
eRegionalHolidayEffect:
Univariate
Resu
lts
Epiphany(1
988-2009)
Epiphany(2
001-2009)
All
Saints’Day(1
988-2009)
All
Saints’Day(2
001-2009)
Corp
usChristi(2
001-2009)
Mark
etCapitalization
0.25***(1
0.30)
0.30***(9
.96)
0.19***(1
0.93)
0.22***(1
1.99)
0.26***(1
4.29)
Idiosy
ncra
ticRisk
-4.23(-1.29)
-3.59(-0.86)
-9.74***(-3.65)
-9.06***(-3.09)
-12.97***(-3.79)
ResidualM
edia
Covera
ge
0.13**(2
.11)
0.07*(1
.85)
0.07*(1
.81)
PanelB:Determ
inants
ofth
eRegionalHolidayEffect:
Multivariate
Resu
lts
Epiphany(1
988-2009)
Epiphany(2
001-2009)
All
Saints’Day(1
988-2009)
All
Saints’Day(2
001-2009)
Corp
usChristi(2
001-2009)
Mark
etCapitalization
0.24***(8
.60)
0.30***(8
.88)
0.16***(8
.69)
0.20***(1
0.24)
0.23***(1
2.92)
Idiosy
ncra
ticRisk
-0.71(-0.20)
-0.56(-0.54)
-9.06**(-2.33)
-8.62***(-2.62)
-7.00**(-1.97)
ResidualM
edia
Covera
ge
0.10*(1
.88)
0.07*(1
.89)
0.05*(1
.78)
Ret +
,t15.96***(7
.11)
15.27***(5
.44)
12.13***(4
.15)
14.57***(8
.50)
19.08***(9
.75)
Ret −
,t17.84***(7
.02)
19.04***(5
.87)
16.11***(8
.07)
16.34***(8
.29)
17.05***(7
.34)
RM
RF
0.12*(1
.74)
0.16*(1
.78)
0.04(0
.92)
0.09(1
.63)
0.13**(2
.17)
HM
LBeta
-0.03(-0.57)
-0.07(-1.29)
0.03(0
.62)
0.05(1
.34)
-0.03(-0.59)
WM
LBeta
-0.04(-0.84)
-0.10(-1.64)
0.03*(1
.95)
0.02(0
.38)
-0.04(-1.17)
Alpha
-0.86(-0.00)
2.16(0
.09)
-2.89(-0.25)
-3.80(-0.27)
17.58(1
.13)
Rura
lDummy
-0.16(-0.95)
0.14(0
.84)
-0.21**(-2.21)
-0.22**(-2.14)
-0.12(-1.25)
52W
eekHigh
Dummy
0.07(0
.62)
0.15(1
.15)
0.20**(2
.22)
0.27**(2
.32)
0.20**(2
.09)
Ad
hocDisclosu
re(t)
0.28(0
.90)
0.80**(2
.47)
0.50*(1
.90)
Media
Covera
ge(t+1)
0.32(1
.46)
-0.11(-1.28)
0.09(0
.88)
Constant
-2.12***(-8.25)
-3.12***(-10.31)
-1.58***(-9.65)
-3.12***(-10.31)
-3.12***(-10.31)
Industry
and
YearDummies
yes
yes
yes
yes
yes
R2
0.18
0.26
0.11
0.16
0.17
N1,997
1,037
3,963
2,422
2,794
38
Tab
le7:
OnlineBroker
Sam
ple:DescriptiveStatistics,
LocalBias,
andTradingShocksduringRegionalHoliday
s
This
table
presents
resu
ltsfrom
theanalysisofretail
investortradingbased
on
dailydata.Thesample
period
startson
January
1,1997and
endson
April17,2001.
Allvalues
are
inEuro.PanelsA
toC
providedescriptivestatistics.
Panel
Dsh
owsth
estandard
ized
fractionoffirm
sth
atare
loca
tedin
thesamemetropolitanareaas
theonlinebroker
across
sample
investorpreference
portfolios.
Investorpreference
forea
chstock
inth
eonlinebroker
sample
isdefi
ned
asth
edifferen
cebetweenafirm
’s
brokerageweightand
itsweightin
themarket
portfolioofGerman
stock
s,averaged
across
month
lyobservations.
Each
portfolioco
ntainsan
equalnumber
ofstock
s.
Values
reported
inpanel
Dare
standard
ized
bydividingea
chfractionbyth
efractionobtained
forth
emed
ium
preference
portfolio.PanelsE
andF
provideresu
ltsfrom
twoseparate
testsofsh
ock
sin
tradingactivityatth
edayofth
eholiday,
asdescribed
indetailin
thetext.
PanelA:Aggregated
Data
(1,084tradingdays,
2,901investors,965Germ
an
stock
s)
No.Tra
nsa
ctions
Tota
lTra
nsa
ction
Value
No.Sells
Tota
lValueofSells
No.Buys
Tota
lValueofBuys
136,125
765,300,000
59,770
381,900,000
76,355
383,300,000
PanelB:Cro
ss-sectionalSta
tisticsforIn
dividualTra
nsa
ctions
Mean
Value
Median
Value
SD
1%
99%
Tra
nsa
ctions:
All
5,622
2,733
16,044
128
44,407
Tra
nsa
ctions:
Buys
5,021
2,523
14,676
125
38,059
Tra
nsa
ction:Sells
6,390
3,066
17,607
140
50,832
PanelC:Cro
ss-S
ectionalSta
tisticsforIn
dividualIn
vestors
Mean
Median
SD
1%
99%
Numberoftransa
ctions
47
22
95
1447
Numberoffirm
straded
15
11
16
185
Valueoftransa
ctions
263,792
65,743
993,767
580
3,288,099
PanelD:Sta
ndard
ized
Fra
ction
ofFirmsLocated
inth
eBro
kerM
etropolita
nAreabySample
InvestorPreferencePortfolios
Low
Preference
Medium
Preference
High
Preference
0.98
11.29
PanelE:PercentilesofRelativeTra
dingActivityduringth
eHolidaybyPreferencePortfolios
All
Portfolios
Low
Preference
Medium
Preference
High
Preference
ValueofTra
nsa
ctions
31
61
42
24
NumberofTra
nsa
ctions
32
49
44
29
NumberofActiveIn
vestors
35
50
43
29
PanelF:PercentilesofShock
Variable
atth
eEventDaybyFirm
Size
SizeGro
up
Epiphany1997
Epiphany1998
Epiphany1999
Epiphany2000
All
Saints’Day1999
All
Saints’Day2000
Corp
usChristi2000
Size:All
50
75
46
43
28
72
32
Size:Larg
e50
72
64
56
16
66
44
Size:Small
34
39
25
16
30
511
39
Appendix for “The Trading Volume Impact of Local Bias:
Evidence from a Natural Experiment”
Heiko Jacobs and Martin Weber∗
June 1, 2011
Abstract
This appendix contains explanations and tables that supplement the analysis in the paper. Table1 gives an overview of the data sets used. Table 2 illustrates the distribution of legally recognizedholidays across German states. Table 3 provides descriptive statistics of the stock market data. Table4 displays the distribution of industry groups across samples. Table 5 illustrates the construction ofthe factors for size, value and momentum. Table 6 provides further evidence on the level of differencesin shocks in absolute abnormal returns between holiday and non-holiday firms. Figure 1 comparesthe cumulative distribution functions of shock variables on Epiphany. Table 7 compares individualinvestor characteristics in holiday and non-holiday regions.
∗Heiko Jacobs is from the Lehrstuhl fur Bankbetriebslehre, Universitat Mannheim, L 5, 2, 68131 Mannheim. E-Mail:
[email protected]. Martin Weber is from the Lehrstuhl fur Bankbetriebslehre, Universitat Mannheim, L
5, 2, 68131 Mannheim, and CEPR, London. E-Mail: [email protected].
1
Tab
le1:
Overview
ofDataSets
Data
Data
Source
Description
Sample
Period
Dailystock
mark
etdata
Thomso
nReuters
Data
stream
Adjusted
and
unadjusted
dailyclosingprices,
mark
etcapitalization,bookvalues,
thenumberofdaily
sharestraded,th
enumberofto
talsh
aresoutsta
nding,ad-
justmentfacto
rs,th
eprimary
exch
angeaswell
asindustry
membership
forafinal
sample
of792stock
soffirm
sheadquartered
inGerm
any
June13,1988-January
15,2009
Daily
trading
record
sof
reta
il
clients
Germ
an
OnlineDiscountBro
ker
316,134stock
transa
ctionsofabout3,000individualinvestors;outofth
ese,136,125
transa
ctionsare
conducted
by2,901investors
with
regard
to965firm
sheadquar-
tered
inGerm
any
January
1,1997-April17,2001
Analyst
covera
ge
I/B/E/S
Recommendation
and
review
dates
for
ato
talof51,497
stock
buy/hold/sell-
recommendationspublish
ed
by196bro
kers,coveringabout80%
ofsa
mple
firm
s
January
1,1993
-December31,
2008
Media
covera
ge
Genios,
Factiva
126,125hand
collected
firm
-specificnewsarticlespublish
ed
inth
reeleadingGer-
man
daily
business
papers
(FinancialTim
esDeutsch
land,Sueddeutsch
eZeitung,
Handelsblatt),
coveringabout94%
ofsa
mple
firm
s
January
1,2000
-January
15,
2009
Adhocdisclosu
resandcorp
ora
te
news
Deutsch
eGesellschaft
fuer
Ad-
hoc-P
ublizitaet(D
GAP)
Tim
e-sta
mped
ad
hoc
disclosu
resforall
sample
firm
s,time-sta
mped
corp
ora
te
newsforhalf
ofsa
mple
firm
s
January
1,2000
-January
15,
2009
Intern
et
search
frequencies
for
firm
names
Sta
ndard
izeddailytime-seriesofsearchfrequenciesforth
enamesofsa
mple
firm
s,
constru
cted
from
Google’s
application
“In
sights
forSearch”
January
1,2004
-January
15,
2009
Indexmembership
Deutsch
eBoerseAG
Tim
eSeriesofhisto
ricalDAX
and
MDAX
indexcomposition
June13,1988-January
15,2009
Individual
investor
chara
cteris-
tics
SAVE
study
Quantita
tiveinform
ation
on
variouseconomic
and
socio-p
sych
ologicalch
ara
cter-
isticsofabro
adly
representa
tivesa
mple
ofGerm
an
individualinvestors
based
on
arich
panelquestionnairestudy
Yearlydata
from
2005-2009
Metropolita
nregions
Initiative
Euro
pean
Metropoli-
tan
Regionsin
Germ
any
Deta
iled
inform
ation
on
thebord
ers
ofth
eeleven
Germ
an
metropolita
nareasas
defined
byth
eConferenceofM
inisters
forSpatialPlanning
Firm
headquarters
Thomso
nReuters
Data
stream,
Deutsch
eBoerseAG,firm
home-
pages
Address
offirm
headquarterlocation
2
Table
2:Legally
Recog
nized
Holiday
sin
GermanStates
This
tablesillustratesth
edistributionoflegallyauth
orizedholidaysacross
Germanstates.
Theillustrationis
constru
cted
onth
ebasisofanoverview
provided
byth
e
Ministryofth
eInterior(see
www.bmi.bund.defordetails).Holidayscovered
inth
eanalysis(E
piphany,
All
Saints’Day,
Corp
usChristi)
are
highlighted
inbold
type.
Note
that,
whileReform
ationDayis
anoth
erregionalholiday(celeb
ratedin
most
ofth
eform
erGermanDem
ocraticRep
ublic),less
than5%
ofsample
firm
sare
loca
ted
inth
eaffectedarea(see
alsofigure
1).
Holiday
Date
BW
BA
BE
BRA
BR
HA
HE
MW
PLS
NRW
RP
SAA
SAX
SAXA
SH
TH
New
Year’sDay
January
1X
XX
XX
XX
XX
XX
XX
XX
X
Epip
hany
January
6X
XX
Good
Friday
EasterSundayminus2days
XX
XX
XX
XX
XX
XX
XX
XX
EasterSunday
X
EasterM
onday
EasterSundayplus1day
XX
XX
XX
XX
XX
XX
XX
XX
MayDay
May1
XX
XX
XX
XX
XX
XX
XX
XX
Ascension
Day
EasterSundayplus39days
XX
XX
XX
XX
XX
XX
XX
XX
Whit
Sunday
EasterSundayplus49days
Whit
Monday
EasterSundayplus50days
XX
XX
XX
XX
XX
XX
XX
XX
CorpusChristi
EasterSunday
plu
s60
days
XX
XX
XX
Augsb
urg
erFriedensfest
August
8A
Assumption
August
15
kX
DayofGerm
an
Unity
Octo
ber3
XX
XX
XX
XX
XX
XX
XX
XX
Reform
ation
Day
Octo
ber31
XX
XX
X
All
Sain
ts’Day
Novem
ber1
XX
XX
X
DayofRepenta
nce
Wednesd
aybefore
November31
X
Christmas
December25and
December26
XX
XX
XX
XX
XX
XX
XX
XX
X=legalholiday
BW
=Baden-W
urttemberg
LS=LowerSaxony
A=only
inAugsb
urg
BA=Bavaria
NRW
=North
Rhine-W
estphalia
k=legalholidayin
predominantlycath
olicmunicipalities
BE=Berlin
RP=Rhineland
Palatinate
BRA=Bra
ndenburg
SAA=Saarland
BRE=Bremen
SAX=Saxony
HA=Hamburg
SAXA=Saxony-A
nhalt
HE=Hesse
SH=Sch
lesw
ig-H
olstein
MW
P=M
eck
lenburg
-Western
Pomera
nia
TH=Thuringia
3
Table
3:SummaryStatisticsBased
onWeekly
Data
This
table
provides
varioussu
mmary
statisticsforoursample
based
on
792German
firm
sin
theperiod
from
June13,1988to
January
15,2009.Panel
Adescribes
cross-sectionaldifferen
cesin
time-series
averages
offirm
characteristics.Firm
market
capitalization
ismea
sured
inmillion
Euro.Firm
book/market
−ratiois
the
balance
sheetvalueofth
eco
mmoneq
uityin
theco
mpany(W
orldscopeitem
03501)divided
byth
emarket
valueofeq
uity.
Firm
turnover
isth
eweekly
number
ofsh
ares
traded
scaledbyth
enumber
oftotalsh
aresoutstanding.Firm
weeksis
thenumber
ofweeksth
efirm
isin
oursample.Panel
Bdescribes
time-series
characteristics
of
weekly
cross-sectional(equally-andvalue-weighted)averages.Numberoffirmsis
thenumber
ofactivefirm
sin
oursample
inagiven
week.
Pan
elA:Cross-section
alStatistics
Variable
Mean
Med
ian
SD
5thPercentile
95th
Percentile
Firm
return
0.09
%0.16
%0.58
%-0.95%
0.67%
Firm
return
volatility
7.33
%6.37
%5.99
%2.97
%12.94
%
Firm
market
capitalization
1,14
812
34,70
417
4,190
Firm
book/market-ratio
0.52
0.44
1.36
0.09
1.04
Firm
turnover
1.42
%0.93
%1.68
%0.01%
4.55
%
Firm
weeks
556
515
322
951,075
Pan
elB:Tim
e-series
Statistics
Variable
Mean
Med
ian
SD
5thPercentile
95th
Percentile
Value-weightedreturn
index
0.16
%0.31
%2.54
%-4.22%
3.96%
Equally-w
eightedreturn
index
0.21
%0.35
%1.73
%-2.69%
2.48%
Value-weigh
tedturnover
index
2.04
%1.86
%0.98
%0.80%
3.69%
Equally-w
eigh
tedturnover
index
0.98
%0.85
%0.56
%0.33
%2.09%
Number
offirm
s41
140
915
118
3620
4
Tab
le4:
Distribution
ofIndustry
GroupsAcrossHoliday
andNon-H
oliday
Regions
This
table
showsth
eco
ncentrationandco
mpositionofindustries
(asgiven
byth
eDatastream
Lev
el2industry
classifica
tion)across
holidayandnon-holidayregionsfor
theca
seofEpiphany,
AllSaints’Day,
andCorp
usChristi,resp
ectively.
Panel
A(B
)sh
owsaneq
ual-weighted(value-weighted)analysis,
whereth
eweightofea
chindustry
groupis
determined
byth
efractionofholidayfirm
sin
thatindustry
group(byth
erelativemarket
capitaliza
tionofindustry
groupobservations).In
Panel
B,DAX
30
bluech
ips(about6%
ofobservations)
are
excluded
topreven
tasm
allfraction
oflargefirm
dominatingth
eanalysis.
Inboth
panels,
theHerfindahlindex
ofindustry
concentrationis
computedasth
esu
mofsquaredindustry
groupweights.
Industry
Epiphany
All
Saints’Day
Corp
usChristi
Holiday
Non-H
oliday
Holiday
Non-H
oliday
Holiday
Non-H
oliday
Tota
lnumberoffirm
s301
491
509
283
603
189
PanelA:Equal-W
eighted
Analysis
Herfi
ndahlindexofidustry
concentration
0.15
0.15
0.15
0.16
0.15
0.15
Oil,Gas,
Altern
ativeEnerg
y3.32%
3.67%
2.75%
4.95%
2.65%
6.35%
BasicM
aterials
5.98%
7.13%
7.27%
5.65%
6.80%
6.35%
Industrials
22.26%
21.59%
22.79%
20.14%
22.39%
20.11%
Consu
merGoods
17.28%
14.66%
16.11%
14.84%
15.92%
14.81%
Health
Care
5.65%
6.11%
4.52%
8.48%
5.31%
7.94%
Consu
merServ
ices
7.97%
8.15%
9.23%
6.01%
7.96%
8.47%
Telecommunications
0.33%
1.02%
0.98%
0.35%
0.83%
0.53%
Utilities
4.32%
2.44%
4.13%
1.41%
3.65%
1.59%
Financials
18.27%
22.20%
18.27%
25.09%
19.90%
23.28%
Tech
nology
14.62%
13.03%
13.95%
13.07%
14.59%
10.58%
PanelB:Value-W
eighted
Analysis
Herfi
ndahlindexofidustry
concentration
0.16
0.16
0.17
0.18
0.17
0.17
Oil,Gas,
Altern
ativeEnerg
y3.37%
3.79%
1.73%
4.79%
1.61%
14.41%
BasicM
aterials
8.56%
9.81%
10.06%
8.13%
10.59%
7.90%
Industrials
20.44%
20.99%
23.25%
19.22%
21.16%
16.32%
Consu
merGoods
14.11%
12.59%
12.31%
10.40%
12.64%
8.24%
Health
Care
3.52%
7.81%
3.06%
12.34%
4.37%
2.22%
Consu
merServ
ices
6.39%
7.22%
7.82%
4.63%
6.92%
7.63%
Telecommunications
1.25%
0.57%
0.93%
0.77%
0.65%
1.47%
Utilities
12.18%
3.53%
8.59%
1.57%
7.74%
3.54%
Financials
25.57%
26.59%
26.54%
31.07%
28.97%
30.97%
Tech
nology
4.61%
7.10%
5.70%
7.08%
5.34%
7.30%
5
Tab
le5:
Constructionan
dSummaryStatisticsof
Factors
forSize,
Value,
andMom
entum
Thefactors
forsize,valueandmomen
tum
are
constru
cted
from
thedailystock
retu
rnsofth
ose
firm
swhichsu
rviveth
escreen
ingprocess
asoutlined
insection3.The
screen
ingprocess
isdesigned
toex
cludevery
small
and
illiquid
stock
sand
toiden
tify
firm
swith
available
and
reliable
stock
market
data.This
procedure
resu
ltsin
afinalsample
of792firm
shea
dquartered
inGermany.
With
regard
toth
eco
nstru
ction
ofvalueand
size
portfolios,
wefollow
veryclosely
themethodologyproposed
inFamaandFrench
(1993).
Specifica
lly,
inea
chyea
ratth
een
dofJune,
weform
sixvalue-weightedportfoliosbasedonfirm
size
andeq
uitybook-to-m
arket
ratio.A
firm
’seq
uitybook-to-m
arket
ratiois
defi
ned
asth
ebalance
sheetvalueofth
eco
mmon
equityin
theco
mpany(W
orldscopeitem
03501)divided
byth
emarket
value
ofeq
uity.
Toqualify
forth
einclusion
inany
ofth
eportfolios,
afirm
hasto
meetth
efollowingrequirem
ents:1)Thestock
must
havevalid
price
data
atth
een
dof
June(i.e.nopreviousdelistingand
notbeingdisca
rded
by
thescreen
ingprocess.2)Thebook
aswellasth
emarket
valueasco
mputed
atth
een
dofDecem
ber
of
thepreviousyea
rhaveto
benon-neg
ative.
To
sort
stock
sinto
portfolios,
weform
threebook-to-m
arket
equity
groups(low,med
ium,high)based
on
the30th
and
70th
percentile
ofth
ebook-to-m
arket-ratioaswellas(indep
enden
tly)tw
osize
equitygroups(small,big)based
on
themed
ian
ofth
emarket
capitaliza
tion.From
the
intersectionsofth
esegroups,
thefollowingsixportfoliosare
constru
cted
:SmallLow,SmallMed
ium,SmallHigh,Big
Low,Big
Med
ium,Big
High.Dailyvalue-weighted
retu
rnsonth
esixportfoliosare
calculatedfrom
thebeg
inningofJuly
toth
een
dofJuneofth
enex
tyea
r.Theportfoliosare
reform
edyea
rly.
Dailyfactorretu
rnsfor
size
(SMB)and
value(H
ML)are
then
calculated
asfollows:
SM
B=
1/3(S
mallLow
+SmallM
edium
+SmallHigh)−
1/3(B
igLow
+BigM
edium
+BigHigh)and
HM
L=
1/2(S
mallHigh+
BigHigh)−
1/2(S
mallLow
+BigLow).
Withregard
toth
emomen
tum
factor,
wefollow
veryclosely
themethodologyproposedin
Carh
art
(1997).
Specifica
lly,
weform
threeeq
ually-w
eightedportfolios,
splitbyth
e30th
and70th
percentile
ofth
edistributionofth
emost
recentelev
en-m
onth
sretu
rnlagged
one
month
(i.e.th
emost
recentmonth
isskipped
).This
resu
ltsin
threemomen
tum-sorted
portfolios(W
inners,
Neu
tral,Losers),
whichare
reform
edmonth
ly.Tobeincluded
inanyofth
eportfolios,
thestock
must
havevalidprice
data
forth
ewhole
previousyea
r(i.e.nopreviousdelistingandnotbeingdisca
rded
byth
escreen
ingprocess).
Dailyretu
rnsforth
emomen
tum
factor(W
ML)are
then
calculatedasfollows:
WM
L=
Win
ner
s−Losers.
Thefollowingtable
showssu
mmary
statisticsoffactorretu
rns
basedonmonth
lydata
from
June1988to
January
2009(236observations).Month
lyfreq
uen
cyis
chosento
facilitate
comparisonwithoth
erstudies.
Thetable
displays
month
lyretu
rnsu
mmary
statisticsforth
emarket
factor(R
MRF),
thesize
factor(S
MB),
thevaluefactor(H
ML)andth
emomen
tum
factor(U
MD).
Themarket
factoris
computedasth
emonth
lyretu
rnofth
eCDAX
minusth
emonth
lyrisk-freerate.TheCDAX
isabroadGermanstock
index
withseveralhundredco
nstituen
ts.Accord
ing
toth
eindex
provider
Deu
tsch
eBoerse
Group,it
reflects
theperform
ance
ofth
eoverallGermaneq
uitymarket.
Facto
rM
ean
Std
.Dev.
t-statistic:mean=0
Min
Max
RM
RF
0.25%
5.79%
0.67
-19.90%
18.63%
SM
B-0.24%
3.50%
-1.05
-13.12%
12.77%
HM
L1.01%
2.86%
5.44
-8.76%
9.90%
WM
L0.68%
4.65%
2.26
-22.77%
14.09%
6
Table 6: Difference in Shocks in Absolute Abnormal Returns: Further Evidence
This table provides supplementary material for the test on differences in shocks in absolute abnormal returns,
as described in section 4.3 of the paper. Panel A reports differences in mean shocks in absolute abnormal
returns. To this end, daily absolute abnormal returns for each firm during t-5 to t+5, as obtained from a
German version of the Carhart (1997) four factor model, are computed. Factor loadings are estimated from
time-series regressions from t-66 to t-6. For both the holiday and the non-holiday sample, firm-specific shocks
are computed as the absolute abnormal return at t minus the average absolute abnormal return in t-5 to t+5
(excluding t). The table reports the difference between the mean shock value for the holiday sample and the
mean shock value for the non-holiday sample, averaged across years. To mitigate the effect of extreme outliers,
we winsorize the data at the 1% and 99% level before computing the mean. This is done for the holiday and
non-holiday sample in each year separately. Statistical significance is assessed by bootstrapping as described in
footnote 12. Large firms (Small firms) refer to stocks with a market value larger (smaller) than the median
stock, measured at the beginning of the year. Statistical significance at the ten, five and one percent level is
indicated by *, **, and ***, respectively. Panel B compares the frequency of extreme return events on the event
day. To this end, all holiday (non-holiday) firm-level shocks are pooled. Shock variable at least 0% means
that the idiosyncratic component of the stock’s return on the event day has at least the same importance as
on average in a nearby benchmark period (t-5 to t+5, excluding t). The odds ratio is computed as the ratio
of the fraction of extreme events for holiday firms and the fraction of extreme events for non-holiday firms.
Panel A: Difference in Mean Shocks in Absolute Abnormal Returns
Dependent Variable Epiphany All Saint’s Day Corpus Christi Pooled
Abnormal absolute return: All firms -0.08%* 0.02% -0.07% -0.04%
Abnormal absolute return: Large firms -0.09%** 0.00% -0.08% -0.06%*
Abnormal absolute return: Small firms -0.07% 0.02% -0.06% -0.03%
Panel B: Frequency of Extreme Return Events on the Event Day
Event Holiday Firm Observations Non-Holiday Firm Observations Odds Ratio
% of Observations with Shock Variable at least 0% 34.57% 37.09% 0.93
% of Observations with Shock Variable at least 1% 15.22% 16.04% 0.95
% of Observations with Shock Variable at least 2% 7.63% 8.68% 0.88
7
Figure 1: Comparison of cumulative distribution functions of shock variables on Epiphany
The following graph is intended to illustrate the economic magnitude of the difference in shock variables
between holiday and non-holiday firms (see section 4.3 of the paper). As the largest difference is observed
for Epiphany (see panel A of table 4 in the paper ), we employ the following procedure. For each year in
which Epiphany falls on a trading day, we compute the empirical cumulative probably distribution of the
shock variable for holiday firms and separately for non-holiday firms. To obtain an overall distribution, we
then average the resulting percentiles across time. This approach resembles the procedure used in the analysis
relied on in the paper, which aimed at obtaining an estimate for the shock variable of the median firm. The
following graph shows the two cumulative distribution functions. For better readability, only values above the
5th percentile and below the 95th percentile are displayed.
8
Tab
le7:
Individual
Investor
Characteristicsin
Holiday
andNon
-Holiday
Regions
Calculationsin
this
table
are
basedondata
from
theGermanSAVE
study,
arich
panel
survey
producedbyth
eMannheim
Resea
rchInstitute
forth
eEco
nomicsofAging.
Itis
designed
toprovidequantitativeinform
ation
on
theeconomic
situ
ation
and
on
relevantsocio-psych
ologicalch
aracteristics
ofarepresentativesample
ofGerman
households.
Foradetailed
descriptionofth
edesignandresu
ltsofth
eSAVE
study,
wereferth
ereader
toBoersch-Supanet
al.(2009).
Findingsreported
inth
efollowing
table
rely
ondata
gath
ered
infiveco
nsecu
tiveyea
rlyquestionnairestudieswithth
esamehouseholds,
conducted
between2005and2009.Thetable
summarizesvarious
characteristics
ofindividualinvestors
inholidayandnon-holidayregionsforea
chofourth
reeholidaysamples(E
piphany,
AllSaints’Day,
Corp
usChristi).If
notstated
oth
erwise,
thevariablesrepresentsimple
mea
nsormed
iansco
nstru
cted
from
thepooled
sample
ofall
yea
rly
survey
s.Relyingon
varioussets
ofweightingfactors
to
reca
librate
thesample
withth
eaim
ofoptimized
representativen
essdoes
notch
angeth
equalitativenatu
reofourfindings.
Theco
mputationofallvariables(w
ithth
e
exceptionof“stock
market
participation”)is
conditioned
onhouseholdswhoinvestin
thestock
market
(e.g.via
individualstock
s,butalsovia
mutu
alfunds,
REIT
Setc.).
InPanel
B,su
bjectsco
uld
givemultiple
resp
onses.
“Nodiscu
ssionoffinancialmatters”
refers
toth
efractionofsu
bjectswhostatedto
rely
neith
eronrelatives,friends,
collea
gues,neighbors
norfinancialadviserswhen
dea
lingwithfinancialmatters.
InPanel
C,th
eex
tentto
whichth
eadviseis
followed
isjudged
onascale
from
0(n
otat
all)to
10(completely).
InPanel
D,“Objectivefinancialknowledge”
reportsth
efractionofparticipants
whoansw
ered
allofth
reefinancialliteracy
questionsco
rrectly.
This
variable
isco
mputedrelyingonly
onth
equestionnaireof2009.Thequestionsth
ereinare
designed
toevaluate
knowledgewithregard
tointerest
rates,
inflationand
risk
ofinvestm
entalternatives,resp
ectively(see
www.m
ea.uni-mannheim.defordetails).“Self-assessedfinancialknowledge”
gives
resp
onden
ts’statemen
tsregard
ingth
eir
subjectivemea
sure
offinancialknowledgeonascale
from
1(verylow)to
7(veryhigh).
InPanel
E,“self-assessedrisk-taking”is
evaluatedbyaskingsu
bjectsto
assessth
e
validityofth
efollowingstatemen
tonascale
from
0(completely
false)
to10(completely
true):“Idonotmindtakingrisksin
investm
ents”.“Expectationswithregard
toownfutu
reeconomic
situ
ation”are
ratedonascale
from
0(veryneg
ative)
to10(verypositive).
9
InvestorCharacteristic
Epiphany
AllSaints’Day
Corp
usChristi
Holiday
Non-H
oliday
Holiday
Non-H
oliday
Holiday
Non-H
oliday
Panel
A:FinancialSituationandParticipationin
theStock
Market
Participationin
thestock
market?
26.35%
24.04%
25.81%
23.51%
25.80%
23.08%
Valueofstock
market
investm
ents
inEuro
(mea
n)
26,365
30,855
30,589
29,369
31,899
27,527
Valueofstock
market
investm
ents
inEuro
(med
ian)
10,000
10,000
10,000
10,000
10,000
10,000
Totalfinancialwea
lthin
Euro
(mea
n)
65,919
68,937
70,819
66,254
74,999
60,235
Totalfinancialwea
lthin
Euro
(med
ian)
36,000
32,849
32,075
34,488
34,000
33,000
Net
month
lyhousehold
inco
mein
Euro
(mea
n)
3,079
2,988
3,099
2,932
3,156
2,828
Net
month
lyhousehold
inco
mein
Euro
(med
ian)
2,800
2,600
2,800
2,550
2,800
2,500
Panel
B:Influen
ceofSocialContacts(E
xcludingProfessionalFinancialAdvisors)onFinancialMatters
Discu
ssionoffinancialmatterswithrelatives?
28.61%
28.22%
28.41%
28.21%
27.17%
29.66%
Discu
ssionoffinancialmatterswithfriends?
23.12%
24.54%
23.18%
25.12%
23.28%
25.41%
Discu
ssionoffinancialmatterswithco
llea
gues?
7.82%
8.00%
7.19%
8.60%
7.16%
8.94%
Discu
ssionoffinancialmatterswithneighbors?
0.64%
1.96%
1.37%
1.94%
1.63%
1.74%
Nodiscu
ssionoffinancialmatters?
27.45%
28.09%
28.91%
27.14%
29.37%
26.25%
Panel
C:Use
ofFinancialAdvice
Discu
ssionoffinancialmatterswithprofessionalfinancialadvisors?
50.14%
51.80%
51.19%
51.67%
50.73%
52.23%
Ifyes:Adviceseek
edatleast
once
amonth
?4.32%
2.65%
3.12%
2.88%
3.35%
2.57%
Ifyes:Adviceseek
edaboutfourtimes
per
yea
r?23.26%
24.51%
24.21%
24.28%
23.61%
25.00%
Ifyes:Adviceseek
edaboutonce
ayea
r?49.51%
50.87%
50.66%
50.54%
50.73%
50.43
Ifyes:Adviceseek
edless
thanonce
ayea
r?22.91%
21.98%
22.01%
22.29%
22.31%
22.00%
Ifyes:Towhatex
tentis
theadvicefollowed
?(m
ean,scale
from
0to
10)
6.04
6.08
5.98
6.14
6.02
6.13
Ifyes:Towhatex
tentis
theadvicefollowed
?(m
edian,scale
from
0to
10)
66
66
66
Panel
D:FinancialLiteracy
Objectivefinancialknowledge(T
opfinancialliteracy
score)
82.08%
78.59%
79.71%
79.10%
80.00%
78.59%
Self-assessedfinancialknowledge(m
ean,scale
from
1to
7)
4.97
4.93
4.89
4.97
4.91
4.97
Self-assessedfinancialknowledge(m
edian,scale
from
1to
7)
55
55
55
Panel
E:Self-AssessedRisk-T
akingin
Investm
ents
andEco
nomic
Expectations
Self-assessedrisk-takingin
investm
ents
(mea
n,scale
from
0to
10)
2.99
3.02
3.04
3.00
3.06
2.97
Self-assessedrisk-takingin
investm
ents
(med
ian,scale
from
0to
10)
22
22
22
Expectationswithregard
toownfutu
reeconomic
situ
ation(m
ean,scale
from
0to
10)
5.65
5.81
5.80
5.72
5.83
5.70
Expectationswithregard
toownfutu
reeconomic
situ
ation(m
edian,scale
from
0to
10)
66
66
66
10